Restructure documentation, validate metadata removal, consolidate assets

docs/cli.md

@@ -0,0 +1,248 @@
+# CLI guide
+
+The command line interface is organized around the type of work you want to do.
+
+```text
+remove-ai-watermarks [OPTIONS] COMMAND [ARGS]
+```
+
+Run `remove-ai-watermarks COMMAND --help` for the complete option list and
+defaults. This page focuses on choosing the right command.
+
+## Inspect an image
+
+```bash
+remove-ai-watermarks identify image.png
+```
+
+`identify` combines supported metadata and pixel signals into one provenance
+report. When no signal is found, it reports the origin as unknown. It does not
+claim the image is clean.
+
+Machine readable output:
+
+```bash
+remove-ai-watermarks identify image.png --json
+```
+
+Metadata only inspection:
+
+```bash
+remove-ai-watermarks identify image.png --no-visible
+```
+
+Despite the historical option name, `--no-visible` skips both visible and open
+invisible pixel detectors. Metadata inspection still runs.
+
+## Remove known visible marks
+
+```bash
+remove-ai-watermarks visible image.png -o clean.png
+```
+
+The default behavior:
+
+- checks every registered visible mark;
+- removes every detected match;
+- selects the best installed fill backend;
+- strips AI metadata from the output.
+
+Use a specific mark:
+
+```bash
+remove-ai-watermarks visible image.png --mark gemini -o clean.png
+```
+
+Available mark names are printed by:
+
+```bash
+remove-ai-watermarks visible --help
+```
+
+Keep metadata:
+
+```bash
+remove-ai-watermarks visible image.png --keep-metadata -o clean.png
+```
+
+Use the strict visual gate without metadata or sibling corroboration:
+
+```bash
+remove-ai-watermarks visible image.png --sensitivity strict -o clean.png
+```
+
+When no known mark is detected, the command does not write a new output. Use
+`erase` if you can identify the affected region yourself.
+
+## Erase a region
+
+```bash
+remove-ai-watermarks erase image.png \
+  --region 1640,1930,400,100 \
+  -o clean.png
+```
+
+The region format is `x,y,width,height`. Repeat `--region` to erase more than
+one box:
+
+```bash
+remove-ai-watermarks erase image.png \
+  --region 20,20,180,60 \
+  --region 1640,1930,400,100 \
+  -o clean.png
+```
+
+Choose the fill backend:
+
+```bash
+remove-ai-watermarks erase image.png \
+  --region 1640,1930,400,100 \
+  --backend migan \
+  -o clean.png
+```
+
+`erase` accepts `cv2`, `migan`, and `lama`. The corresponding optional extra
+must be installed for a learned backend.
+
+## Strip AI metadata
+
+Inspect metadata:
+
+```bash
+remove-ai-watermarks metadata image.png --check
+```
+
+Remove AI metadata and write a new file:
+
+```bash
+remove-ai-watermarks metadata image.png --remove -o clean.png
+```
+
+When `-o` is omitted, removal overwrites the source. Standard metadata is kept
+unless you pass `--remove-all`.
+
+The command also supports the audio and video containers listed in
+[supported signals](supported-signals.md). ffmpeg must be available for the
+non-ISOBMFF audio and video path.
+
+## Remove invisible watermarks
+
+Install the diffusion dependencies first:
+
+```bash
+uv tool install --force "remove-ai-watermarks[gpu]"
+```
+
+Then run:
+
+```bash
+remove-ai-watermarks invisible image.png -o clean.png
+```
+
+The command normally skips regeneration when no supported local signal is
+detected. Use `--force` when you know the image should be processed:
+
+```bash
+remove-ai-watermarks invisible image.png -o clean.png --force
+```
+
+### Choose a pipeline
+
+| Pipeline | When to use it |
+| --- | --- |
+| `controlnet` | Default compatibility profile with structural conditioning |
+| `sdxl` | Lighter plain SDXL regeneration |
+| `qwen` | Large CUDA oriented Qwen Image profile |
+| `qwen-zimage` | CUDA only high fidelity profile with a separate face stage |
+
+Example:
+
+```bash
+remove-ai-watermarks invisible image.png -o clean.png \
+  --pipeline qwen-zimage --force
+```
+
+The legacy `default` value is an alias for `sdxl`. The `--auto` option is
+deprecated, emits a warning, and changes nothing.
+
+### Work with limited memory
+
+Lower CUDA memory pressure:
+
+```bash
+remove-ai-watermarks invisible image.png -o clean.png \
+  --cpu-offload --force
+```
+
+Keep large images at native resolution while processing them in overlapping
+tiles:
+
+```bash
+remove-ai-watermarks invisible image.png -o clean.png \
+  --tile --max-resolution 0 --force
+```
+
+Or set a resolution cap:
+
+```bash
+remove-ai-watermarks invisible image.png -o clean.png \
+  --max-resolution 2048 --force
+```
+
+Tiling avoids the explicit downscale but each tile is regenerated separately.
+It is a memory strategy, not a guarantee of better quality.
+
+## Run the full pipeline
+
+```bash
+remove-ai-watermarks all image.png -o clean.png
+```
+
+The command runs:
+
+1. visible mark removal;
+2. invisible watermark removal when available and applicable;
+3. AI metadata stripping.
+
+The visible options and diffusion options are also available on `all`.
+
+If diffusion is required but the `gpu` extra is unavailable, `all` still
+writes the result of the visible and metadata stages, prints a prominent
+warning, and exits with code 1. This prevents a partial result from being
+reported as complete.
+
+## Process a directory
+
+```bash
+remove-ai-watermarks batch ./images --mode visible
+```
+
+Modes:
+
+- `visible`;
+- `invisible`;
+- `metadata`;
+- `all`.
+
+Set an output directory:
+
+```bash
+remove-ai-watermarks batch ./images \
+  --mode all \
+  --output-dir ./clean
+```
+
+The invisible and full modes accept the same main diffusion controls as their
+single image counterparts. Run `batch --help` for the authoritative option
+list.
+
+## Exit behavior
+
+The CLI uses nonzero exit codes for meaningful incomplete outcomes, including
+no detected target on commands that would otherwise regenerate or create a
+misleading unchanged result, processing errors, and a required invisible step
+that could not run.
+
+Scripts should check the process exit code and the output path. The detailed
+per-command contract is maintained in
+[module internals](module-internals.md#command-line-interface).

docs/controlnet-removal-pipeline-research.md

@@ -1,5 +1,10 @@
 # ControlNet-as-removal-pipeline research: can structure-conditioned regeneration scrub SynthID and keep text?
 
+> Research archive. This document records experiments, superseded defaults, and
+> deployment considerations from the time of the study. It does not define the
+> current CLI or Python API. See `README.md`, `docs/cli.md`, and
+> `docs/known-limitations.md` for current behavior.
+
 Date: 2026-06-02. Source: a manual primary-source pass (WebSearch + WebFetch over the
 watermark-removal-attack and SDXL-ControlNet literature). Prompted by issue #35
 (@newideas99 / Jacob): "as we use SDXL even at low strength that kills small text ... Do you
@@ -109,11 +114,11 @@ Gemini app; the two payloads are vendor-specific and never cross-checked):
 - **OpenAI 0.20 transfers to prod as-is** (OpenAI removal is resolution-independent:
   the study clears it at 0.05 across 1024-1600).
 - **Gemini 0.30 is the floor at <= 1536 only.** Gemini is resolution-sensitive (study:
-  native 2816 likely needs >= 0.30 even on `default`), and **raiw.cc runs NATIVE**
-  (`max_resolution=0` in `modal_app.py`). So either CAP Gemini to <= 1536 in raiw.cc and
+  native 2816 likely needs >= 0.30 even on `default`). A native-resolution
+  deployment should either cap Gemini to <= 1536 and
   use 0.30, or run a native-resolution Gemini cert and expect a higher floor (~0.35+).
 
-### Recommendations for a removal pipeline (raiw.cc)
+### Recommendations for a removal pipeline
 
 - **Treat controlnet as PRESERVATION, not removal.** Choose it for text/structure content,
   `default` for photoreal; removal efficacy comes from STRENGTH in both.
@@ -205,7 +210,9 @@ on each tile. This mirrors the `_run_region_hires` insight (text needs MORE pixe
 regeneration so strokes exceed the VAE's ~8 px latent floor), but ctrlregen runs the regeneration
 at LOW res, the opposite. CtrlRegen's paper gives no resolution/tiling spec to contradict this.
 
-**Sources.** internal (`src/remove_ai_watermarks/noai/ctrlregen/engine.py`); resolution-omission
+**Sources.** the former internal
+`src/remove_ai_watermarks/noai/ctrlregen/engine.py` (removed after this study);
+resolution-omission
 confirmed against https://arxiv.org/html/2410.05470v1
 
 ### Finding 5 — confidence: high
@@ -287,7 +294,7 @@ ControlNet + activations in **fp32** (MPS fp16 decodes to all-black NaN — issu
 on run 1 below; fp32 is the required default on mps/cpu) — fits the 32 GB budget with vae-tiling +
 attention-slicing; ~1-2 min/image, so a coarse sweep is a sub-hour background run. A dedicated GPU
 is needed ONLY for the separate
-native-large-Gemini (2816 px) case, which OOMs even without a ControlNet (that stays a raiw.cc
+native-large-Gemini (2816 px) case, which OOMs even without a ControlNet (that requires a
 GPU task). The genuine external dependency is NOT compute but the **manual SynthID oracle**:
 there is no local SynthID detector, so removal is verified by hand in the Gemini app
 ("Verify with SynthID") per image, regardless of where the diffusion runs.

docs/index.md

@@ -0,0 +1,38 @@
+# Documentation
+
+This documentation is split by purpose. Start with the user guides if you want
+to run the tool. Use the maintainer references only when changing the code.
+
+## User guides
+
+| Page | Use it when |
+| --- | --- |
+| [Installation](installation.md) | You need the CLI, an optional model backend, or a development environment. |
+| [CLI guide](cli.md) | You want a command for one image, a directory, or a specific watermark type. |
+| [Python API](python-api.md) | You want to call the package from Python. |
+| [Supported signals](supported-signals.md) | You need to know which visible marks, metadata formats, and invisible signals are covered. |
+| [Known limitations](known-limitations.md) | You need the quality, device, format, or verification boundaries. |
+| [Scope, safety, and legal notes](legal-and-safety.md) | You need the intended use and legal context. |
+
+## Maintainer references
+
+| Page | Purpose |
+| --- | --- |
+| [Module internals](module-internals.md) | Current architecture, invariants, and regression guards by module. |
+| [Verification plan](verification-plan.md) | Verification methods, completed measurements, and remaining validation gaps. |
+| [Release and distribution](release-and-distribution.md) | PyPI, Homebrew, Hugging Face Space, and release workflow. |
+| [Watermarking landscape](watermarking-landscape.md) | Vendor signals and detection approaches. |
+| [SynthID technical reference](synthid.md) | Mechanism, detector access, robustness, and implications for this project. |
+
+## Research archive
+
+These pages record experiments and the evidence behind past decisions. They are
+not command references and may describe prototypes that were later removed.
+The current behavior is defined by the code, tests, README, and user guides.
+
+- [ControlNet removal research](controlnet-removal-pipeline-research.md)
+- [Qwen improvement research](qwen-improvement-research.md)
+- [Doubao reverse-alpha research](research-doubao-distillation.md)
+- [SynthID identity research](synthid-robust-identity-research.md)
+- [SynthID identity follow-up](synthid-robust-identity-research-2026-06-08.md)
+- [Text protection research](text-protection-research.md)

docs/installation.md

@@ -0,0 +1,136 @@
+# Installation
+
+Python 3.10.1 or newer is required.
+
+## Core install
+
+The core package provides:
+
+- provenance inspection;
+- visible watermark removal with OpenCV;
+- manual region erasing with OpenCV;
+- AI metadata inspection and removal.
+
+Install it as an isolated command with uv:
+
+```bash
+uv tool install remove-ai-watermarks
+```
+
+Or with pipx:
+
+```bash
+pipx install remove-ai-watermarks
+```
+
+You can also install the Homebrew package on macOS or Linux:
+
+```bash
+brew install wiltodelta/tap/remove-ai-watermarks
+```
+
+## Invisible watermark removal
+
+Diffusion based removal needs the `gpu` extra:
+
+```bash
+uv tool install --force "remove-ai-watermarks[gpu]"
+```
+
+The code supports CUDA, XPU, MPS, and CPU devices. A GPU is recommended because
+CPU inference is slow.
+
+For the CUDA only Qwen Image plus Z-Image profile:
+
+```bash
+uv tool install --force "remove-ai-watermarks[qwen-zimage]"
+```
+
+The `qwen-zimage` extra includes the normal `gpu` dependencies.
+
+## Optional features
+
+Install only what you need:
+
+| Extra | Adds |
+| --- | --- |
+| `migan` | MI-GAN ONNX fill backend |
+| `lama` | big-LaMa ONNX fill backend |
+| `detect` | Open DWT-DCT watermark decoder used by `identify` |
+| `trustmark` | Adobe TrustMark decoder |
+| `esrgan` | Real-ESRGAN upscaling before diffusion |
+| `qwen-zimage` | CUDA only Qwen Image plus Z-Image pipeline |
+
+Example:
+
+```bash
+uv tool install --force "remove-ai-watermarks[migan,detect]"
+```
+
+Some optional models download their weights on first use.
+
+## Install from the repository
+
+```bash
+git clone https://github.com/wiltodelta/remove-ai-watermarks.git
+cd remove-ai-watermarks
+uv sync --frozen
+```
+
+Add the feature groups required for your work:
+
+```bash
+uv sync --frozen --extra dev
+uv sync --frozen --extra dev --extra gpu
+```
+
+Run commands from the repository root:
+
+```bash
+uv run remove-ai-watermarks --help
+```
+
+## Development setup
+
+Install development dependencies:
+
+```bash
+uv sync --frozen --extra dev
+```
+
+Run the complete project gate:
+
+```bash
+bash maintain.sh
+```
+
+The script runs dependency checks, linting, formatting checks, type checking,
+and the test suite.
+
+## Hugging Face authentication
+
+Pass a Hugging Face token directly when the selected model or account requires
+one:
+
+```bash
+remove-ai-watermarks invisible image.png --hf-token "$HF_TOKEN"
+```
+
+The CLI also loads `HF_TOKEN` from the environment and from a local `.env`
+file. The same name is documented in `.env.example`.
+
+## Troubleshooting
+
+### The first model run is slow
+
+Diffusion and learned fill backends may download model weights on first use.
+Later runs reuse their caches.
+
+### The command skips invisible removal
+
+The normal behavior is to skip diffusion when no supported local signal is
+found. A missing signal does not prove that the image is clean. If you know the
+image came from a relevant generator, use `--force`.
+
+If the CLI reports that diffusion dependencies are unavailable, install the
+`gpu` extra.

docs/known-limitations.md

@@ -1,212 +1,213 @@
-# Known limitations: full detail
+# Known limitations
 
-> Relocated verbatim from `CLAUDE.md` on 2026-06-11 to keep the always-loaded
-> context small. Long single-line entries were reformatted into paragraphs;
-> no content was changed or summarized.
+This page describes current product limits. Historical measurements and
+superseded experiments live in the research archive listed in
+[the documentation index](index.md).
 
-Full detail behind the compact Known-limitations list in `CLAUDE.md`:
-measurements, incident history, oracle runs, and the reasoning behind each
-decision. Read the relevant section here before changing the diffusion
-pipelines, strength defaults, or metadata coverage.
+## Visible removal
 
-## Visible-mark fill quality is background- and backend-dependent
+### Fill quality depends on the background
 
-Visible-mark removal is localize -> fill (the reverse-alpha pixel recovery was
-dropped; see `docs/module-internals.md`). The fill only touches the mark's
-footprint, so there is never collateral damage outside it, and whether the mark
-is *removed* is fill-independent -- cv2, MI-GAN and LaMa all strip the mark's
-shape. What varies is the *quality of the recovered region*, and it depends on
-the background:
+Visible removal changes only the selected mask, but the hidden pixels still
+have to be reconstructed.
 
-- **Flat backgrounds:** every backend is clean; cv2 is often the crispest.
-- **Textured / regular-structured backgrounds** (fabric, foliage, a lattice or
-  grid): an inpaint can only guess the hidden pixels. `cv2` (the classical
-  no-deps floor) visibly smears; `migan` (light, learned) can leave a ghost or
-  hallucinate structure; `lama` (heavy, learned) is the most reliable and
-  recovers structure best.
+- OpenCV is fast and dependency free. It works well on flat backgrounds but
+  can smear texture or repeated structure.
+- MI-GAN is a lighter learned backend. It can improve natural texture but may
+  ghost or invent structure.
+- LaMa is the heaviest learned backend and is generally the strongest option
+  for difficult backgrounds.
 
-The old reverse-alpha recovered the *true* pixels under a well-captured, static
-mark, so on structured backgrounds it was sometimes cleaner than any inpaint.
-The trade for localize -> fill is robustness (it also handles moved / re-rendered
-marks and needs no per-mark alpha capture) and a simpler, swappable pipeline.
-`auto` resolves best-first (`LaMa > MI-GAN > cv2`) and warns once when it falls
-back to cv2 because no learned backend is installed; a memory-tight deployment
-that cannot afford LaMa's ~4.7 GB peak pins `--backend migan` explicitly.
+`--backend auto` selects LaMa when available, then MI-GAN, then OpenCV.
 
-## Invisible-pipeline resolution handling (native / 1024 floor / `--max-resolution`; MPS memory tiers)
+No backend can recover detail that is completely hidden by an opaque mark. A
+successful detection therefore does not guarantee a visually perfect fill.
 
-`invisible` pipeline processes at **native resolution for inputs whose long side is >= 1024px**, and **auto-upscales smaller inputs UP to a 1024px floor** (`min_resolution=1024`, the default; `--min-resolution 0` disables) before diffusion -- SDXL img2img distorts badly on a tiny latent (a 381x512 portrait wrecks at native, the #36 follow-up), and the output is restored to the original input size so the floor is a transparent quality boost (it adds time/memory on small inputs). The floor upscale uses Lanczos by default; **`--upscaler esrgan`** (opt-in, the `esrgan` extra) runs Real-ESRGAN first for better detail before the Lanczos resize to the exact target (`upscaler.py` / `InvisibleEngine._esrgan_upscale`, falls back to Lanczos if the extra is absent). `max_resolution=0` (default) means no downscale cap, matching the hosted raiw.cc backend (fal fast-sdxl, no pre-downscale). The old forced downscale-to-1024 -> upscale-back round-trip for LARGE images was the main quality loss (issue #10) and is gone; at strength ~0.05 SDXL img2img does not need a downscale.
+### Automatic detection covers registered variants only
 
-**CUDA model CPU offload (`--cpu-offload`):** Diffusers normally places the complete SDXL, ControlNet, or base Qwen pipeline on the selected CUDA device. The opt-in offload mode instead uses Diffusers model-level CPU offload, keeping one model component on the GPU at a time and moving it back to CPU before the next component runs. This reduces peak VRAM use but adds transfer overhead. The custom `qwen-zimage` runtime already manages its global model placement; the same flag forces its face stack to stay on the offload path instead of becoming resident on a high-memory GPU. CPU and MPS behavior is unchanged.
+The registry contains vendor and locale specific templates. A redesigned mark,
+an unsupported locale, a different position, or a crop may be missed.
 
-**Final `--unsharp` post-filter (`humanizer.unsharp_mask`, opt-in, default 0):** applied LAST (after the face-restore pass, else it would be smoothed over) to counter the soft/over-smoothed look diffusion + restoration leave (an AI tell); ~0.5-0.8 safe, higher risks halos. Pairs with `--humanize` (grain adds sensor-noise texture, unsharp adds crispness). `--max-resolution N` re-introduces an opt-in long-side cap purely to bound GPU/MPS memory on very large inputs (it reintroduces the lossy round-trip). For huge images that OOM at native, **`--tile` is the lossless alternative** -- see the tiled-diffusion subsection below.
+Known examples:
 
-### Tiled diffusion for large inputs (`--tile`, issue #10)
+- Samsung detection is calibrated for the Italian
+  `Contenuti generati dall'AI` text variant.
+- The Jimeng top-left pill has a weak visual detector and is intentionally
+  subject to additional product and background checks.
+- Kling support covers the calibrated variants rather than every Kling label.
 
-`--tile` (OFF by default; `--tile-size` default 1024, `--tile-overlap` default 128) processes the diffusion pass in overlapping sliding-window tiles instead of one forward pass, so a large image is regenerated at **native resolution** without the OOM and without the lossy `--max-resolution` downscale round-trip. It engages only when the long side exceeds `--tile-size`; a sub-tile image runs a single pass unchanged. The SDXL, ControlNet, and base Qwen paths refactor the single-image `_generate` into a per-tile `_generate_one` (the ControlNet canny edge map is rebuilt per tile, so structure preservation works tile-local) and route it through `noai.tiling.run_tiled` when tiling is active. `qwen-zimage` instead tiles only its global Qwen pass, feather-blends that result, and then runs YuNet, SAM, and Z-Image once against the full original/result pair. The geometry and blend math are pure helpers, unit-tested without the model (`tests/test_tiling.py`):
+Use `erase --region` when you can see and select an unsupported or missed mark.
 
-- `plan_tiles(w, h, tile_size, overlap)` lays out a row-major grid where every tile is exactly `tile_size` (the last tile on each axis is pulled back flush to the far edge, simply overlapping its predecessor more). Uniform tile size keeps each diffusion pass at SDXL's preferred dimension.
-- `feather_weights(w, h, overlap)` is a separable linear taper, ~1 in the interior and ramping toward each edge, kept **strictly positive** so the normalized accumulate-and-divide blend (`accum / weight_sum`) is a partition of unity: a region covered by one feathered edge (an image corner) still divides cleanly. Identical (unchanged) tiles therefore reconstruct the input exactly -- the seam-free guarantee, asserted in `test_identity_generate_reconstructs_image`.
+### Strict and automatic sensitivity trade recall for precision
 
-CAVEAT: each tile is an **independent** low-strength regeneration. At the current SDXL/ControlNet defaults (0.10-0.15) the per-tile drift is small and the feather blend hides the seams, but tiling is a memory workaround, not a quality upgrade over a single native pass -- a 32 GB MPS box that clears the native UNet peak should prefer no tiling. The MPS->CPU fallback still applies per tile; if the first tile falls back to CPU, the device stays CPU for the rest of the image.
+`--sensitivity strict` uses the visual gate alone. The default `auto` mode may
+relax a mark only when metadata or a confidently detected sibling mark
+corroborates the same product.
 
-For `qwen-zimage`, the global denoise is still computed from the full-frame megapixel count and the same resolved seed is reused for every tile. The profile defaults to seed 0, matching the release-candidate oracle run; an explicit seed overrides it. Running the face stage only after blending avoids duplicate regeneration and boundary-local face misses. A real H100 smoke on 2026-07-25 exercised the shipped branch on a 4096x3072 input (20 tiles at 1024 with 128 px overlap, seed 0, strength 0.154): it completed in 653.367 seconds after 43.741 seconds of setup, preserved the exact dimensions, and peaked at 22.732 GiB allocated / 23.861 GiB reserved CUDA memory. Visual inspection found no tile seams. The worst tile-boundary gradient-change line was at the 98.563 percentile of all image lines (2.522 standard deviations), below the preselected 99th-percentile outlier threshold; overview fidelity was MAE 3.332%, PSNR 26.564 dB, and global SSIM 0.988627. This no-face input validates the global tiled execution and blend, not the post-blend face path. The July 25 seed-0 oracle result still certifies exact non-tiled candidate bytes only; tiled SynthID efficacy requires a separate provider-oracle check.
+There is no blanket "this image is AI" relaxation. That information does not
+identify the vendor, mark, or location and caused unacceptable false
+detections in the removed experimental mode.
 
-**Concrete MPS data points (the OOM is memory-tier-dependent, NOT a hard MPS limit):** on a ~24 GB unified-memory machine (verified 2026-05-25, 1254x1254 gpt-image SDXL, fp32) native res OOMs at the *UNet* step (peak ~17 GiB), not only the VAE decode, and the auto-fallback in `img2img_runner` reloads on CPU and finishes (slow, ~13 min) -- the output is still weight-identical and defeats SynthID, so "looks hung/crashed" on Mac is usually this CPU fallback, not a pipeline error. On a **32 GB** unified-memory machine the same default SDXL pass runs entirely on MPS with **no CPU fallback** (verified 2026-05-31, 1122x1402 gpt-image, `all`/default, ~155 s end-to-end), so 32 GB clears the native-res UNet peak that 24 GB could not. Adding `enable_vae_tiling()` alone does NOT prevent the 24 GB OOM (the peak is the UNet, not the VAE). The fast Mac workarounds for memory-constrained machines are fp16 on MPS (roughly halves memory) or `--max-resolution` to cap the long side; neither is wired as the default. The `controlnet` pipeline adds the canny ControlNet weights on top of SDXL, so its peak is a bit higher than the plain `default` pass; the same MPS->CPU fallback covers an OOM. The native-vs-cap-vs-floor decision lives in the pure helper `invisible_engine._target_size(w, h, max_resolution, min_resolution)` (returns `None` for native, a target tuple for a downscale cap OR an upscale floor; cap takes precedence, the floor is skipped on a min>max misconfig) so it is unit-tested (`tests/test_invisible_engine.py::TestTargetSize`, the #10/#15/#36 regression guard) without loading the model -- keep that logic in the helper, don't re-inline it.
+## Invisible removal
 
-## fp16 VAE black-output fix (issue #29) + degenerate-output fp32 backstop (issue #41)
+### Regeneration is lossy
 
-**fp16 VAE black-output fix (issue #29, 2026-05-30):** on a **CUDA/XPU fp16** backend the stock SDXL VAE overflows to NaN and the *plain* img2img path decodes to an **all-black** image (reproduced on the raiw.cc result: a 1086x1448 input -> a uniformly black 4.6 KB PNG, mean 0). `watermark_remover._load_pipeline` / `_load_controlnet_pipeline` swap in the fp16-fixed SDXL VAE (`madebyollin/sdxl-vae-fp16-fix` = `_SDXL_FP16_VAE_ID`) when `_needs_fp16_vae_fix(model_id, DEFAULT_MODEL_ID, is_fp16)` is true -- only the default SDXL checkpoint on fp16.
+Invisible removal does not decode and delete a payload. It regenerates the
+image through a diffusion pipeline. Faces, text, colors, and fine detail can
+change even when the watermark is successfully disrupted.
 
-**cpu/mps run fp32** (the stock VAE is fine there, which is why the bug never reproduces on Mac). A custom non-SDXL `model_id` keeps its own VAE (the fp16-fix VAE is SDXL-architecture-specific). The decision is a pure helper, unit-tested without a download (`tests/test_platform.py::TestFp16VaeFix`); the actual black->clean recovery needs a CUDA GPU.
+ControlNet is the default compatibility profile. It conditions on edges to
+preserve structure, but edges do not preserve identity or exact texture.
 
-**Confirmed on real CUDA hardware 2026-06-03:** running `all` on a 1086x1448 OpenAI gpt-image (the #29 repro size) at fp16 produced a normal (non-black) output, so the fp16-fix VAE swap resolves the all-black decode. (It was not reproducible on this MPS machine, which runs fp32, so the verification had to happen on an NVIDIA box.)
+The CUDA only `qwen-zimage` profile adds a separate face stage and is the
+highest fidelity option in the current implementation. It is larger, slower,
+and still may alter small text or difficult faces.
 
-**Follow-up safety net (issue #41, 2026-06-04):** the swap is gated to `model_id == DEFAULT_MODEL_ID`, so a custom model, a stale pre-fix install, or a fal/custom loader can still hit the black decode -- a new reporter did (gpt-image 1448x1086, the #29 size, with the exact `image_processor.py:142 invalid value encountered in cast` warning the NaN->0 cast emits). `remove_watermark` now adds a model-agnostic backstop: after generation, if the run was fp16 AND the output is degenerate (`_is_degenerate_image`: mean and std both below `_DEGENERATE_THRESHOLD` 1.0 -- a uniform all-black/NaN frame; the variance guard spares a legitimately dark-but-textured photo), it rebuilds the pipeline in fp32 on the SAME device and re-runs once. fp32 is the verified-clean path, so the user never gets a black image regardless of model_id/version. Mirrors the existing MPS->CPU fallback's self-mutation pattern (reset `torch_dtype` + clear `_pipeline`/`_controlnet_pipeline`); `batch` inherits it through `remove_watermark`, and once one image trips it the rest of the batch stays on the safe fp32. The detector is a pure helper, unit-tested without a model (`tests/test_platform.py::TestDegenerateOutputGuard`); the full fp16->detect->fp32-retry chain was verified e2e on this MPS machine by forcing fp16 with the swap disabled (first pass black, guard fired, retry produced a normal image). CAVEAT: the fp32 retry uses ~2x memory, so on a VRAM-constrained GPU it can OOM (a visible error, still better than a silent black frame; the MPS->CPU fallback covers that path). The reporter's "CPU also black" symptom is NOT reproducible here -- fp32 (cpu/mps) decodes clean -- so it points at an old version or a non-fp32 run, pending their version + command.
+### Removal cannot be verified locally for proprietary SynthID
 
-## rich was dropped (plain-text CLI and scripts)
+The project has no public local SynthID pixel decoder. It can infer likely
+presence from supported provenance metadata, but a missing metadata proxy is
+not a negative pixel verdict.
 
-**rich was dropped (CLI + scripts print plain text via `click.echo`).**
+For important outputs:
 
-`cli.py` renders through small `_Console`/`_Table`/`_Progress` shims; the analysis scripts (`scripts/synthid_corpus.py`, `synthid_pixel_probe.py`, `text_detection_benchmark.py`, `corpus_gap_scan.py`) import `Console`/`Table` from the shared `scripts/_plain_console.py` shim (markup like `[bold]`/`[/]` is stripped, tables render aligned). Consequences: (1) `rich` is NOT a dependency, so anything that imports it breaks a clean `uv sync --frozen` (CI installs core+dev only) — this exact gap red-failed CI after the refactor when those 4 scripts still imported rich; if you add a script, use the `_plain_console` shim, not rich. (2) The old `[gpu]`-bracket-eaten bug (#19) is gone — plain `click.echo` prints `pip install 'remove-ai-watermarks[gpu]'` verbatim, no escaping needed (regression-guarded by `tests/test_cli.py::TestGpuHintMarkup`). (3) No Unicode glyphs / colors / progress bars in CLI output by design.
+1. preserve the original;
+2. process a copy;
+3. verify with the matching provider tool when available;
+4. do not assume one provider's verifier covers another provider's payload.
 
-## AVIF/HEIF/JPEG-XL metadata, ISOBMFF/ffmpeg removal, audio watermark detection
+Provider systems can change, so a result verified on one file, seed, or version
+is not a permanent certification.
 
-Metadata detection for AVIF/HEIF/JPEG-XL relies on a binary scan for `C2PA_UUID` + `IPTC_AI_MARKERS`, plus EXIF `Software` / XMP `CreatorTool` generator tags via `metadata.exif_generator` (validated with synthesized AVIF/JPEG fixtures + an XMP raw-scan fixture). C2PA removal in those containers is implemented via `noai/isobmff.py` (top-level ``uuid`` / ``jumb`` box stripper, no re-encoding), which now also drops a top-level XMP ``uuid`` box that carries an AI label (matched by AI-marker content, not by the XMP UUID, so byte-order-robust) and covers MP4/MOV/M4V/M4A by content sniff.
+### Strength is content and seed dependent
 
-**Non-ISOBMFF audio/video removal is via ffmpeg** (`_FFMPEG_STRIP_EXTS` -> `_strip_with_ffmpeg`): WebM/Matroska (EBML), MP3 (ID3), WAV/FLAC/OGG (RIFF/Vorbis) are stripped losslessly with `ffmpeg -map_metadata -1 -map_chapters -1 -c copy` (codec data untouched). Requires ffmpeg on PATH; raises `RuntimeError` if absent or if ffmpeg can't parse the file. Verified end-to-end (a real ffmpeg-made WAV/MP3 with a `title=Suno AI` tag -> tag gone, audio bytes preserved).
+For SDXL and ControlNet, the CLI resolves an unset strength from the detected
+vendor:
 
-**Meta-box XMP now handled (`isobmff.blank_ai_xmp_packets`, v0.6.9):** an AI-label XMP packet stored as a meta-box `mime` item (AVIF/HEIF) is blanked in place (overwritten with spaces of the same length, so `iloc` offsets and the coded image stay valid).
+- OpenAI: `0.10`;
+- Google: `0.15`;
+- unknown: `0.15`.
 
-**`Exif` item inside the `meta` box (AVIF/HEIF), now handled in place (2026-06-19):** an AI-generator token in an EXIF item (its TIFF bytes live in `mdat`/`idat`) is blanked by `isobmff.blank_ai_exif_tokens` — it finds EXIF TIFF blocks by their II/MM byte-order header, validates each with **piexif** (a coincidental II/MM run in pixel data won't parse as a TIFF IFD, so it is ignored), and overwrites any `Software`/`Make`/`Artist`/`ImageDescription` value carrying an `AI_GENERATOR_TOKENS` token with spaces of the **same length**. Same-length means every box size and `iloc` offset stays valid and the coded image is untouched — so it avoids the full `iinf`/`iloc` surgery (offset rewrite) that exiftool would need (exiftool is a non-installed binary dep, deliberately not used). It scrubs only the AI value; camera/editor EXIF is preserved. Wired into `remove_ai_metadata`'s ISOBMFF path after `blank_ai_xmp_packets`. Because the ISOBMFF branch never runs the JPEG `_scrub_ai_exif`, this is the ONLY EXIF scrubber on that path and must stay in PARITY with it: it now also blanks the China TC260 `{"AIGC":{...}}` block in `ImageDescription`/`UserComment` (via `_is_aigc_exif_value` — Doubao producer + Tencent service-provider schemas) and the xAI/Grok `Signature:` + UUID-`Artist` pair, not just `AI_GENERATOR_TOKENS` in `Software`/`Make`/`Artist`/`ImageDescription` (regression `test_noai.py::TestISOBMFF::{test_blank_aigc_block_in_exif, test_blank_xai_signature_pair_in_exif}`). **Still NOT built:** Resemble PerTh audio detection (no presence/confidence flag exists).
+An explicit `--strength` overrides these defaults. The defaults are operating
+points, not universal guarantees. Near a removal threshold, different content
+or a different random seed may change the verifier result.
 
-**Audio watermark DETECTION (Resemble PerTh) was evaluated and NOT built (2026-05-26):** `resemble-perth`'s `PerthImplicitWatermarker.get_watermark()` returns a raw bit-array with **no presence/confidence flag** (clean audio decodes to arbitrary bits too), so reliably distinguishing watermarked-from-clean needs either Resemble's fixed payload or a confidence API -- neither is public, and there's no real Resemble sample to calibrate against. Same wall-class as the SynthID pixel detector: the decode exists, reliable presence-detection does not. (perth's top-level `PerthImplicitWatermarker` is also gated to None unless `librosa` is importable.)
+The base Qwen and `qwen-zimage` profiles have profile specific strength
+behavior. Consult `remove-ai-watermarks invisible --help` and the source of
+[`watermark_profiles.py`](../src/remove_ai_watermarks/noai/watermark_profiles.py)
+for the current resolver.
 
-## SynthID detection is metadata-only (no local pixel detector)
+### Pipelines have different quality tradeoffs
 
-**SynthID detection is metadata-only.**
+| Pipeline | Main limit |
+| --- | --- |
+| `controlnet` | Edge conditioning can preserve a watermark carrying region too closely, and faces may drift. |
+| `sdxl` | Flat graphics and precise structure may receive too little or unhelpful change. |
+| `qwen` | Large CUDA oriented model; face smoothing can still be significant. |
+| `qwen-zimage` | CUDA only, large model stack, and limited broad certification across seeds and content. |
 
-There is no reliable *local* detector of the SynthID *pixel* watermark — Google's decoder is proprietary, no public spec or API (only a waitlisted portal). Authoritative confirmation: Google DeepMind's own paper "SynthID-Image: Image watermarking at internet scale" (Gowal et al., arXiv:2510.09263) states the verification service is restricted to "trusted testers" and does not release detector weights or a reproducible algorithm — so a local pixel detector is infeasible by design, not just unbuilt. https://arxiv.org/abs/2510.09263 We detect SynthID by its C2PA companion (`synthid_source` / `SYNTHID_C2PA_ISSUERS`), which is reliable while the manifest is intact but says nothing once C2PA is stripped.
+The legacy `default` profile name maps to `sdxl`. The `--auto` flag is
+deprecated, emits a warning, and changes nothing.
 
-**Surface-dependent blind spot (verified 2026-05-24):** the same Google model emits different metadata per surface -- the Gemini *app* wraps outputs in Google C2PA, but the *API/playground* (AI Studio, Nano Banana / gemini-2.5-flash-image) emits the SynthID *pixel* watermark (confirmed via the Gemini-app oracle) + the visible sparkle but **no C2PA/IPTC at all**, so `synthid_source` returns None despite SynthID being present. Only the pixel oracle or the visible-sparkle detector catches those. (Meta AI is another surface mismatch: it writes the IPTC `digitalSourceType=trainedAlgorithmicMedia` marker, not C2PA and not SynthID.) Google→SynthID is long-standing; OpenAI→SynthID is confirmed by OpenAI's Help Center (ChatGPT/Codex/API "include both C2PA metadata and SynthID watermarks", updated 2026-05-21) but time-gated (pre-rollout OpenAI images carry C2PA without SynthID), so the OpenAI verdict is hedged "likely". Oracles: Gemini app "Verify with SynthID" (Google), openai.com/verify (OpenAI).
+## Resolution and memory
 
-**Each vendor's oracle detects only its OWN content (verified on the page 2026-05-31):** `openai.com/research/verify` states verbatim "OpenAI generation signals will only be detected if the image was generated with our tools" and "Content could also still be AI-generated by another company's model, which the tool currently does not detect" -- SynthID is shared tech but the verifier is keyed to its own vendor's payload, so a Google-SynthID image reads clean on OpenAI's verifier and vice-versa.
+### Small images are enlarged before SDXL based diffusion
 
-**This explains the recurring "oracle says clean but `identify` still flags SynthID" report (#14):** the oracle reads the *pixel* watermark (gone after our SDXL pass), while `identify` reads the *C2PA-metadata proxy* (still present if the manifest survived). Different signals, not a contradiction -- strip the metadata too (`metadata --remove` / `all`) and the proxy goes quiet, but a quiet proxy is not proof the pixel watermark is gone.
+The SDXL, ControlNet, and base Qwen paths use a default minimum long side of
+`1024`. Smaller inputs are enlarged before diffusion and restored to their
+original dimensions afterward. Set `--min-resolution 0` to disable the floor.
 
-**Consequence for the P0#5 no-signal skip (`has_invisible_target`, 2026-06-22):** `invisible`/`all`/`batch` skip the diffusion scrub by default when no invisible AI signal is *locally* detectable, to avoid degrading a clean image (`--force` overrides). Because SynthID detection is metadata-only, a real AI image whose C2PA was **already stripped** (e.g. a re-encoded download, or the API/playground surfaces above that never emit C2PA) reads as no-signal and is therefore **skipped** — leaving its pixel SynthID in place. This is the deliberate trade: the skip's message never claims the image is clean, and the user re-runs with `--force` when they know it is AI. The blind spot is the same metadata-only ceiling, not a new bug; the visible-sparkle path (`check_visible`) still catches the no-C2PA Gemini-playground case for the *visible* mark, but not the invisible one.
+`qwen-zimage` does not apply this SDXL minimum resolution floor.
 
-**SynthID is durable to JPEG re-encode by design, so a GitHub-recompressed issue attachment is still a valid SynthID test subject** (verified 2026-06-01 on issue #14's pic3: the GitHub-served JPEG survived re-encoding and openai.com/verify still detected SynthID). Do NOT dismiss issue-attachment JPEGs as "not faithful originals" when reproducing a SynthID-survival report: the recompression strips the **C2PA metadata** (so `identify` reads Unknown on the attachment) but NOT the **pixel watermark** that openai.com/verify reads. A true byte-original only matters for the metadata/C2PA path, not for the pixel-SynthID-removal test. (Contrast the open imwatermark above, which IS fragile to JPEG.) The spectral phase-coherence approach from `github.com/aloshdenny/reverse-SynthID` was evaluated (May 2026) and **does not work for real-content detection**: on its own shipped codebook + validation set, watermarked and cleaned images were indistinguishable (conf within noise, cleaned often higher); it only fires on pure-black 1024x1024 reference images at exact resolution (the controlled case it was calibrated on). The README's "90% / conf=0.91" reproduces only in that lab condition. Do not build a production detector on it; if revisited, it is experimental/diagnostic only and needs a per-resolution, per-model reference corpus. A from-scratch gpt-image pilot (2026-05-24) confirmed this independently: 5 independent solid-black gpt-image outputs share a near-identical fixed signature (pairwise residual correlation **0.92**, avg-template retains 97% energy), so the watermark/carrier IS strongly present and consistent on flat content — but the carrier frequencies extracted from it do NOT discriminate real content (carrier-to-random ratio: cleaned 1.86 > watermarked 1.53; a non-gpt-image image scored highest at 3.67). The signature drowns in content texture. Net: a perfectly consistent solid-color signature still yields no real-content pixel detector with magnitude/carrier methods. A corpus discrimination test (2026-05-24, `scripts/synthid_pixel_probe.py`, raw zero-mean residual NCC) independently re-confirms this: at matched resolution, SynthID positives do NOT cluster apart from negatives (within-Gemini 0.07; at 1024 px pos-vs-neg >= pos-vs-pos). The only high correlations were near-duplicate *content* (5 ChatGPT renders of one prompt at ~0.92, while a distinct ChatGPT image scored ~0 against them) — content, not a carrier. The probe is solid-fills-only and EXPERIMENTAL/DIAGNOSTIC; do not use it on real content.
+### Large images stay at native resolution unless capped
 
-**Correction (deeper re-examination 2026-05-25):** the carrier IS real on solid fills — the earlier "no carrier" was a *method* artifact of using spatial / FFT-magnitude NCC, which can't see it. The carrier is a fixed *phase* at specific low frequencies, so the right metric is **per-bin phase coherence**. On 8 white `gemini-2.5-flash-image` fills (generated via the reverse-SynthID trick: identity-edit prompt "Recreate this image exactly as it is" on a synthetic pure-white PNG — this bypasses the recitation block that rejects text prompts for pure colors), phase coherence at the white carriers `(0,±7..±12,±20..±23)` = **0.86** vs **0.31** random; single-image leave-one-out phase-match **+0.83** vs real photos **-0.24**. (Black `2.5-flash` fills clip to std≈0 — SynthID can't push values below 0, so no carrier in black; the repo's dark carriers come from nano-banana-pro.)
+`--max-resolution 0` means no explicit downscale cap. A positive value caps the
+long side before diffusion and restores the result afterward. This reduces
+memory use but introduces a downscale and upscale round trip.
 
-**But it does not generalize:** (a) carriers are model-version + resolution + color specific — the repo's v4 codebook (built for `gemini-3.1-flash-image-preview` + `nano-banana-pro-preview`) scores ~0.527 on my 2.5-flash white fills, indistinguishable from negatives (~0.50), i.e. carriers shift across model versions and need a per-model codebook; (b) on real content (30 `2.5-flash` images) the carrier collapses — set phase coherence at carriers 0.37 ≈ random 0.42, and the repo's v4 detector gives content 0.518 ≈ negatives 0.504 (no separation; a faint +0.24 single-image lean is likely a brightness confound). Net: the spectral/phase approach is a real *controlled-fill* characterizer, NOT an arbitrary-real-content detector, and is brittle to model version. Metadata proxy + visible sparkle + online oracles remain the ceiling for real content.
+`--tile` preserves the input dimensions while running the diffusion stage in
+overlapping tiles. It avoids the explicit downscale, but it is not pixel
+lossless: each tile is independently regenerated. With `qwen-zimage`, only the
+global Qwen stage is tiled; the face stage runs after tile blending.
 
-## External AI-vs-real classifier models are out of scope
+### CPU offload is CUDA only
 
-**External AI-vs-real classifier models are out of scope (decided 2026-05-24).**
+`--cpu-offload` moves Diffusers model components between CPU and CUDA instead of
+keeping the complete standard pipeline in GPU memory. For `qwen-zimage`, it
+forces the face stack to use its offload path.
 
-Generic HuggingFace detectors (`Organika/sdxl-detector` Swin Transformer, `umm-maybe/AI-image-detector`, and fine-tunes) exist and report ~0.98 on their *own* SDXL-vs-real validation sets, but they are per-generator and the model cards themselves note degraded accuracy off-distribution; they are untested on gpt-image / Gemini Nano Banana (the metadata-stripped surfaces we care about), and our own light SDXL pass would likely defeat them the same way it defeats SynthID. Detection here stays local + signal-based (metadata + visible sparkle); do not add a bundled classifier dependency.
+The option reduces CUDA memory pressure at the cost of speed. It has no effect
+on CPU or MPS and fails loudly when a CUDA Diffusers pipeline does not expose
+the required offload method.
 
-## Default strength is vendor-adaptive, one ladder for both pipelines
+### MPS may fall back to CPU
 
-**DEFAULT STRENGTH IS VENDOR-ADAPTIVE, ONE LADDER FOR BOTH PIPELINES (LOWERED 2026-06-14; raised + unified 2026-06-09; vendor-adaptive since 2026-06-01, SUPERSEDES every fixed-default claim in this bullet and the next).**
+The SDXL paths include an MPS out-of-memory fallback that reloads on CPU. A run
+that appears much slower after an MPS failure may be continuing on CPU.
 
-`resolve_strength(strength, vendor)` + `vendor_for_strength(path)` (`watermark_profiles.py`) read the C2PA issuer (`metadata.synthid_source`) on the ORIGINAL input and pick `OPENAI_STRENGTH` **0.10** / `GEMINI_STRENGTH` **0.15** / `UNKNOWN_STRENGTH` **0.15** when `--strength` is unset; explicit `--strength` always wins.
+Memory needs depend on the pipeline, input size, dtype, and machine. Use tiling,
+a resolution cap, or a lighter pipeline when necessary.
 
-**The SAME ladder applies to BOTH pipelines** (`sdxl` and `controlnet`). **2026-06-14: lowered from the 2026-06-04 cert floors (OpenAI 0.20 / Google 0.30) back toward the original 2026-06-01 study (OpenAI ~0.05-0.10 / Google 0.15).** A re-test on the deployed Modal controlnet worker cleared SynthID on the oracle at OpenAI 0.10 (2 photoreal, 1402/1448 px) and Google 0.15 (2 NATIVE 2816x1536 images -- retiring the "native ~2816 likely needs >=0.30" guess), while a pixel sweep showed 0.20/0.30 over-regenerated for no efficacy gain (Google MAE -20% at 0.15). See `watermark_profiles.py` "Data basis". **CAVEATS that stand:** (1) removal near this floor is SEED-NON-DETERMINISTIC (the 2026-06-09 finding below) -- a SERVICE on this ladder must pin a fixed, oracle-verified seed, not rely on a random one; (2) the re-test is n=2 per vendor on photoreal/landscape, NOT flat graphics (the `sdxl` weak spot), so raise `--strength` if an oracle reads SynthID on a flat output.
+## Metadata and formats
 
-**Why one ladder (NOT a per-pipeline split):** the cert was run on controlnet and does NOT transfer to `sdxl` by symmetry (opposite hard cases -- controlnet leaves SynthID on photoreal, `sdxl` on flat graphics), BUT on its OWN hard case (flat fills) `sdxl` is the WEAKER remover (plain img2img barely perturbs a flat region at low strength), so it needs AT LEAST controlnet's strength -- hence the certified floor is the right floor for `sdxl` too. It is a MARGIN argument for `sdxl`, not a fresh certification (no local SynthID detector to self-verify); raise `--strength` if an oracle still reads a flat `sdxl` output. The higher strength costs little quality because `controlnet` is now the default pipeline AND the only `--auto` pick, so `sdxl` is reached only via an explicit `--pipeline sdxl` (a deliberate opt-down for inputs without faces/text), where over-regeneration has nothing to damage. (A short-lived per-pipeline split ladder -- `sdxl` 0.15/0.20 vs controlnet 0.20/0.30 -- existed on 2026-06-09 before being unified the same day; the `resolve_strength` `pipeline` param and the `CONTROLNET_*_STRENGTH` constants were removed.) The CLI detects the vendor from the pristine source (before the visible pass / metadata-strip removes C2PA from the temp file) and passes it to display calls so display and execution agree; `cmd_invisible`/`cmd_all`/`batch` thread `vendor`.
+### Missing metadata does not mean clean
 
-**This replaces the single 0.30 default AND the prior "do NOT build a vendor-adaptive default" policy** -- both came from the now-debunked region-rescrub-contaminated study (the per-region re-scrub that contaminated those numbers was removed in the controlnet refactor). Basis: the oracle-verified June 2026 controlled study (clean v0.8.6, protect OFF): OpenAI clears at 0.05 across 1024-1600 (n=4, resolution-independent); Google needs 0.15 on the capped-1536 path (n=4). `docs/synthid.md` §2.2 (data) + §5.2 (the adaptive default) are authoritative.
+Screenshots, social platforms, and re-encoding can remove metadata while a
+pixel watermark remains. `identify` therefore reports unknown rather than
+clean when no supported signal is found.
 
-**CAVEAT (oracle pass 2026-06-04): the OpenAI 0.10 default is content-dependent, NOT universal -- a flat-graphic OpenAI logo/poster still read SynthID-detected after `default` at 0.10, and photoreal images after controlnet at 0.10/0.15 (low-change regions under-perturbed). Removal at 0.10/0.15 is content×pipeline dependent (see the controlnet Known-limitations bullet); the lever is a higher strength, oracle-revalidated per content type. Do NOT assume the vendor-adaptive default clears every image.**
+### JPEG XL is metadata only
 
-CAVEAT: Google's 0.15 was validated only on `--max-resolution 1536`; native large Gemini (2816) was not locally measurable (OOM on M-series) and is pending GPU validation on raiw.cc -- if it survives 0.15 native, raise `--strength`.
+The metadata path recognizes JPEG XL containers, but the visible and diffusion
+image paths do not list `.jxl` as a supported pixel format because the package
+does not include a JPEG XL pixel decoder.
 
-**Everything below in this bullet about a fixed 0.10/0.30 default is HISTORICAL; trust the vendor-adaptive constants + docs/synthid.md.**
+### HEIC, HEIF, and AVIF use a Pillow fallback
 
-## SynthID removal: strength + oracle scope
+OpenCV does not decode these formats in the project. `image_io.imread` falls
+back to Pillow with `pillow-heif`. A corrupt or truncated file may still fail to
+decode.
 
-**SynthID removal: strength + oracle scope.**
+### Some metadata removal requires ffmpeg
 
-Default strength is vendor-adaptive (see the bullet above); `docs/synthid.md` §2.2 is authoritative for the numbers.
+WebM, Matroska, MP3, WAV, FLAC, OGG, Opus, and AAC container metadata is stripped
+through ffmpeg with stream copying. The operation fails if ffmpeg is absent or
+cannot parse the input.
 
-**Oracle scope (load-bearing):** the Gemini app "Verify with SynthID" is the ONLY valid SynthID oracle (detects Google's mark on any image); `openai.com/verify` is scoped to OpenAI provenance (its own C2PA), NOT a SynthID oracle -- a negative there is meaningless for SynthID. There is no local SynthID detector, so the tool cannot self-check; if the oracle still reads SynthID, raise `--strength` to the lowest value that verifies clean. The profiles are `sdxl` (plain SDXL img2img; `default` is a back-compat alias), `controlnet` (SDXL + canny ControlNet), `qwen` (Qwen-Image img2img), and the experimental `qwen-zimage` two-stage stack.
+### Metadata transformation is fail safe
 
-**Forensic-stealth caveat** (arXiv:2605.09203): defeating the SynthID verifier is NOT forensic invisibility -- independent detectors flag *removal-processed* images vs genuinely-clean ones at >98% TPR@1%FPR, so do not over-claim "indistinguishable from a real photo".
+`remove_ai_metadata` may copy an undecodable file through unchanged instead of
+raising. User facing callers must use `strip_and_verify` and inspect its
+surviving marker mapping before reporting success. The CLI does this.
 
-## `controlnet` pipeline: content x pipeline removal, certified floors, no face-restore
+### Sixteen bit PNG output is not preserved
 
-**`controlnet` pipeline (text/face STRUCTURE preservation, THE DEFAULT since 2026-06-09; `--pipeline default` opts down to plain SDXL).**
+The Pillow based PNG metadata rewrite uses the normal image save path and may
+reduce a sixteen bit PNG to eight bits. A byte-level PNG metadata stripper
+would be required to preserve that bit depth.
 
-SDXL + the canny ControlNet `xinsir/controlnet-canny-sdxl-1.0` via `StableDiffusionXLControlNetImg2ImgPipeline` (`watermark_remover._run_controlnet` / `_load_controlnet_pipeline`).
+## Detection extras
 
-**Removal still comes from the img2img regeneration (`strength`); the ControlNet only PRESERVES text and face STRUCTURE by conditioning on the canny edge map** (`cv2.Canny(gray, 100, 200)`, 3-channel). Canny preserves edges, NOT face identity (a regenerated face drifts in likeness). The drifted cleaned face is the LEAST-AI state we can reach without re-introducing SynthID; **the library does NOT ship a face-restore extra** (every approach evaluated 2026-06-04 - 2026-06-08 -- GFPGAN-on-cleaned, PhotoMaker-V2, InstantID txt2img, InstantID img2img-on-cleaned at three parameter sweeps -- regenerated the face via SDXL and made it look MORE AI-generated). Full empirical conclusion in `docs/synthid-robust-identity-research-2026-06-08.md` "Empirical follow-up". For production face preservation, ship the cleaned image as-is. No original pixels are copied or frozen, **BUT removal at the low vendor-adaptive strength is CONTENT × PIPELINE dependent and NEITHER pipeline clears all content -- oracle-validated against the OpenAI verifier 2026-06-04 (8 images, strength 0.10/0.15, `--max-resolution 1536`).**
+The `detect` extra decodes an open DWT-DCT watermark used in some Stable
+Diffusion, SDXL, and FLUX workflows. That decoder is sensitive to the carrier
+and transformations. A negative result is not a universal negative.
 
-The survivors FLIP by content type: **photoreal** (a 9-face grid, a bracelet product photo) SURVIVES controlnet but CLEARS `default` (controlnet's dense edge map keeps the regen too close to the original, so the SynthID-destroying perturbation never happens; plain img2img perturbs photoreal texture enough); **flat graphic** (a logo/poster with large flat color fills) SURVIVES `default` but CLEARS controlnet (at low strength img2img barely changes flat fills so SynthID persists there, while controlnet repaints them more freely); a flat **text** card cleared under both.
+The `trustmark` extra adds Adobe TrustMark decoding. The implementation retains
+an additional JPEG re-encode gate because isolated decoder hits can otherwise
+be content noise.
 
-**Root cause is insufficient STRENGTH, not the pipeline: at 0.10 the low-change regions -- dense-edge photoreal under controlnet, large flat fills under `default` -- are not perturbed enough to destroy SynthID. The vendor-adaptive 0.10 from the June study is NOT universally sufficient (that study's content happened to clear at 0.10).**
+External AI versus real image classifiers are out of scope. The project
+identifies concrete local provenance signals instead of shipping a generic
+statistical classifier.
 
-The robust fix is a HIGHER strength, oracle-revalidated per content type (controlnet can be cranked harder without losing structure; a lower `controlnet_conditioning_scale` also frees the regen on photoreal). So at today's default strength **both pipelines AND `--auto` can LEAVE SynthID on some content** -- a removal-priority caller (raiw.cc) MUST oracle-validate strength across content types before adopting, not pick a pipeline and assume removal.
+## Output and traceability
 
-**Follow-up same day: re-running the two photoreal survivors through controlnet at an explicit `--strength 0.15` cleared BOTH on the oracle -- BUT one of them (the bracelet) had SURVIVED the SAME 0.15 controlnet config in the first pass (only the random, unset seed differed). So removal near the threshold is SEED-NON-DETERMINISTIC: the same image+pipeline+strength+resolution can pass or fail run-to-run (img2img uses `seed=None`/random unless `--seed` is passed, and there is no local SynthID detector to self-verify). 0.15 is the borderline, NOT a robust floor -- pick a strength with MARGIN (controlnet ~>= 0.20) rather than exactly on it; the content×pipeline table's 0.15 data point is near-threshold noise. A confirming run at `--strength 0.20` controlnet cleared BOTH photoreal survivors on the oracle (ladder: 0.10 grid detected → 0.15 borderline/non-deterministic → 0.20 both clean), so **0.20 is the recommended robust controlnet floor for OpenAI photoreal** (one margin run, not an N-run repeatability proof -- a service should add margin or verify repeatability since there is no local SynthID detector to self-check).
+Removing file-local signals does not remove:
 
-**Engineering follow-up DONE 2026-06-09 (three coupled changes):** (1) **strength raised + unified** -- `resolve_strength(strength, vendor)` now applies ONE vendor-adaptive ladder (the certified controlnet floors 0.20/0.30/0.30) to BOTH pipelines; see the DEFAULT STRENGTH bullet above for why one ladder covers `sdxl`. (2) **`controlnet` is now the DEFAULT pipeline** (CLI `--pipeline` default = `controlnet` + both engine ctors). Rationale: with the certified higher ladder it clears BOTH content classes that flipped in the content-x-pipeline table (photoreal AND flat graphic), whereas plain SDXL left SynthID on flat graphics -- so controlnet is the more removal-robust default. Cost: every non-`--auto` run now downloads the canny ControlNet weights + a higher memory peak (MPS->CPU fallback covers OOM). (3) **the plain-SDXL profile was renamed `default` -> `sdxl`** (`watermark_profiles.SDXL_PROFILE`/`normalize_profile`); `default` stays as a back-compat CLI/ctor alias (the `--pipeline` Choice accepts `sdxl`/`controlnet`/`default`, a click callback `_normalize_pipeline` maps `default`->`sdxl` AND warns that `default` is deprecated). (4) **the content-detection layer + `--auto` planner were removed and `--auto` was retired to a deprecated alias for `--adaptive-polish`** -- see the dedicated `auto_config.py`-removal bullet above (controlnet is the default pipeline and the polish self-gates, so detection changed nothing). A production caller still needs its own per-vendor/content calibration at its deployed native resolution. The Gemini-native resolution caveat stands: controlnet 0.30 is certified only <=1536.** **CERTIFIED 2026-06-04 via an isolated Modal certification harness, restore OFF, ≤1536, each vendor on its own oracle: controlnet floors are OpenAI 0.20 (2 photoreal × 3 seeds = 6/6 clean; the 0.15-flipper is seed-robust at 0.20) and Gemini 0.30 (0.20 detected → 0.30 clean on 2/2 seeds). OpenAI 0.20 transfers to production (resolution-independent); Gemini 0.30 holds only ≤1536 — Gemini is resolution-sensitive, so a native-resolution caller should cap Gemini to ≤1536 at 0.30 or calibrate its native path (~0.35+). Production recipe: controlnet + per-vendor floor in `resolve_strength` (not the default ladder) + FIXED seed (kills the non-determinism).
+- provider account history;
+- server side copies or provenance stores;
+- perceptual fingerprints;
+- evidence that an image passed through a removal pipeline;
+- legal disclosure duties.
 
-**No face-restore runs in the default controlnet profile:** every earlier approach evaluated there (GFPGAN-on-cleaned, PhotoMaker-V2, InstantID txt2img, InstantID img2img-on-cleaned, 2026-06-04 - 2026-06-08 cert sweeps) regenerated the face via SDXL diffusion -- the output face inherited SDXL "clean skin" gloss and lost original identity precision, looking MORE AI-generated than the cleaned image, not less. The separate experimental `qwen-zimage` profile now tests a different architecture, Z-Image regeneration from the original SAM-masked face crop. Its first ArcFace/LPIPS run is recorded below, but it still needs its own oracle and multi-image face/text matrix.**
-
-See `docs/synthid.md` §5.5 + `docs/controlnet-removal-pipeline-research.md` (certified floors table).** **Lesson: visual-quality + face-recovery validation does NOT prove watermark removal -- only the SynthID oracle does, across MULTIPLE content types; never infer removal from sharpness/identity, and never conclude from a partial result (the photoreal-only data first read as "controlnet shields, default removes" -- the flat-graphic result reversed it).**
-
-`controlnet_conditioning_scale` (CLI `--controlnet-scale`, default 1.0) is the structure-preservation knob (higher = closer to the original structure); fp32 on cpu/mps, fp16-fixed VAE on cuda/xpu. The `controlnet` profile is threaded explicitly (`WatermarkRemover(pipeline=...)` / `InvisibleEngine(pipeline=...)`), NOT inferred from `model_id`. This productionizes the `scripts/controlnet_sweep.py` prototype; see `docs/controlnet-removal-pipeline-research.md`.
-
-**Forensic-stealth caveat still applies** (arXiv:2605.09203): defeating the SynthID verifier is not forensic invisibility -- a "this image went through a removal pipeline" classifier can still flag the output.
-
-## `qwen` pipeline (experimental, Qwen-Image 20B, certified floors)
-
-`--pipeline qwen` runs `QwenImageImg2ImgPipeline` on `Qwen/Qwen-Image` (20B MMDiT, Apache-2.0 code AND weights), as an img2img alternative to the SDXL pipelines. Motivation: the controlnet over-regeneration problem above (it plasticizes real photos / loses fine text at the scrub floor). Qwen-Image renders text natively (incl. CJK) and preserves structure markedly better, so at the strength that removes SynthID it damages real content far less.
-
-The scrub still comes from the img2img `strength` (same lever as SDXL); the call shape lives in the pure `_build_qwen_kwargs` (uses Qwen's `true_cfg_scale`, not SDXL's `guidance_scale` — the CLI `--guidance-scale` maps onto it, and ~4.0 is typical vs the SDXL default 7.5). bf16 on CUDA. It is **CUDA/cloud-class — the 20B does not fit MPS — so `_run_qwen` has NO MPS→CPU fallback** (unlike the SDXL paths). Cost on Modal A100-80GB is ~$0.05-0.10/image vs SDXL.
-
-**Certified oracle floors (Modal A100-80GB, 2026-06-20):** on native-resolution OpenAI and Gemini cert inputs (`data/qwen_in/`, both controls SynthID-POSITIVE): **OpenAI 0.10** (0.05 and 0.075 still detected; 0.10 clean and SEED-ROBUST — clean on seeds 0-4, so a random seed is safe) and **Gemini 0.25** (0.20 still detected, 0.25 clean on both images; lowered from the 0.30 first measured). Gemini seed-repeat is single-seed (seed 0): the Gemini oracle rate-limits volume, so PIN a seed in production rather than relying on seed-robustness there.
-
-**Fidelity vs controlnet was MEASURED, not eyeballed (`scripts/fidelity_metrics.py`, text scored against a vision-transcribed ground truth in `data/qwen_in/ground_truth.json` + PaddleOCR on the variants; an initial eyeball read was wrong and overturned by the metrics).** Methodology rule: only compare fidelity at each pipeline's OWN oracle-confirmed scrub floor -- i.e. between outputs where SynthID is actually removed in BOTH (controlnet OpenAI 0.10 / Gemini 0.15; Qwen OpenAI 0.10 / Gemini 0.25). An equal-strength comparison is invalid where it leaves one pipeline un-scrubbed (Qwen at 0.15 does NOT clear Gemini SynthID, so that run was dropped). At those scrub floors:
-- **Text:** Qwen wins on substantial Latin/mixed-script text -- OCR CER, controlnet vs Qwen: openai_1 (EN+RU+ZH, both 0.10) 0.385 vs **0.241**, openai_2 (EN, both 0.10) 0.341 vs **0.290**. On a SHORT CJK sign (gemini_1, cnet 0.15 / Qwen 0.25) it is a TIE (0.037 vs 0.037 -- both near-perfect; the earlier Qwen 0.000 was at the higher 0.30, not the certified floor).
-- **Faces:** controlnet wins -- gemini_3, 18 faces (cnet 0.15 / Qwen 0.25): ArcFace identity 0.546 vs 0.382, Laplacian-variance retention 0.62 vs 0.40, face LPIPS 0.09 vs 0.17 (Qwen smooths faces MORE; the gap narrows vs Qwen 0.30 but controlnet still wins clearly).
-
-**Conclusion: Qwen wins TEXT only for clean body text on a plain background with NO faces; controlnet wins faces AND display/decorative text in a scene. So `qwen` is a MANUAL `--pipeline qwen` opt-in, not a routed lane.** A content `--pipeline auto` router + a faces+text mixed dual-pass were prototyped and DROPPED (2026-06-20): on the canonical faces+text case (the abba poster, faces + display text) controlnet won EVERY metric incl. text (CER 0.114 vs qwen 0.379), so grafting qwen text only hurts; and "text→qwen" is undecidable cheaply (body-vs-display text is what matters). Caveat: `resolve_strength(..., pipeline="qwen")` carries the Qwen ladder (`_QWEN_VENDOR_STRENGTH`, Gemini 0.25), so `--pipeline qwen` gets the 0.25 Gemini floor automatically — the old manual `--strength 0.25` workaround is retired. `_build_qwen_kwargs` now passes an explicit height/width (qwen squished non-square inputs to 1024² without it). Flat-graphic content was not in the sample.
-
-**Improving Qwen (ship vs improve):** the cited research lives in `docs/qwen-improvement-research.md` -- read it before extending the `qwen` pipeline. Verdict: shippable as an opt-in text lane. **The "add a Qwen-Image ControlNet to fix face smoothing" lead was built, measured, and CLOSED (2026-06-20):** a DiffSynth-Studio Qwen + Apache-2.0 blockwise-canny ControlNet at the Gemini floor 0.25 did NOT restore face skin texture (face Laplacian-variance retention flat 0.40 -> 0.40, 13/16 faces within +-0.02; the SDXL+canny target 0.62 was not approached), because canny carries edges not skin grain and Qwen's higher Gemini floor (0.25 vs SDXL+canny 0.15) forces more smoothing -- and a deep-research sweep confirmed NO permissively-licensed Qwen tile/detail/realism/skin ControlNet exists anywhere (every Qwen conditioning is geometry). So **base Qwen stays the text lane, not a face fix.** The distinct Z-Image face-crop lead is now implemented as `qwen-zimage`; direct face comparisons are below, and its exact current six-output candidate is negative in the corresponding provider oracles. Broad seeded removal and text behavior remain unmeasured. Non-regenerative high-frequency detail re-injection is NOT safe by assumption (the "clean-output high frequencies do not carry the watermark" claim was refuted) -- it must be oracle-gated.
-
-**Seed as a quality lever (measured, openai_1 at 0.10, seeds 0-4):** the seed barely moves whole-image fidelity (img LPIPS 0.062-0.065, SSIM 0.855-0.857, PSNR 28.5-28.7 — flat) but does shift TEXT legibility (OCR CER 0.241-0.290, ~17% spread) -- the seed changes WHICH details get regenerated, not the overall level. So a per-image best-of-N-seed selection is a WEAK, text-only lever (pick the lowest-CER seed that still scrubs; fidelity selection needs no oracle). Not worth the N× cost for general use -- pin one decent seed in prod; reserve best-of-N for text-heavy premium cases.
-
-## `qwen-zimage` pipeline
-
-`--pipeline qwen-zimage` is the recommended high-quality SynthID removal mode when CUDA capacity is available and fidelity matters more than latency or cost. It remains a manual opt-in so the broadly compatible, much cheaper ControlNet path can stay the default. The profile ports the upstream two-stage workflow: an input-resolution Qwen-Image-2512 Lightning Canny pass regenerates the frame, then original face crops are segmented and regenerated with Z-Image Turbo before a feathered paste. DiffSynth requires both pixel inputs and the requested dimensions to use the same /16 latent grid, so each stage makes that small alignment resize internally and restores the global result to the original dimensions. The profile defaults to deterministic seed 0 because the release-candidate oracle evidence was produced at that seed; explicit callers can still override it.
-
-The port is architectural, not bit-identical. The active graph was traced from upstream commit `3007d0351596ae0a78b7074dae7ad179710b1e48`, including its linked Impact Pack implementation. It confirms that the active face path is YOLO + SAM; the MediaPipe node visible on the canvas is unconnected. The port keeps the two adaptive-denoise formulas, four-step Qwen Lightning stage, Canny thresholds and scale, AuraFlow shift 3 equivalent, original-image face source, SAM center + box prompts, IoU-0.93 proposal union with highest-score fallback, detector-box intersection, crop factor 2.5, 768 face guide, 1024 crop cap, eight-step face stage, and paste feather 10.
-
-Four runtime differences remain. This package uses full safetensors instead of the source graph's quantized GGUF models, YuNet instead of Ultralytics YOLO to avoid an AGPL runtime, DiffSynth FlowMatch samplers instead of ComfyUI's DPM++ 2M / SGM Uniform and `res_2s` / `bong_tangent` pairs, and no latent-space 20 px detailer noise-mask feather. The face crop is regenerated in full, then only the feathered SAM pixels are composited back, so generated pixels outside that mask are discarded. These differences prevent an exact-output claim even though the architecture and active decision path match.
-
-The default full-frame denoise is resolution-adaptive, not vendor-adaptive. The face denoise is separate and scales from the largest detected face. `--strength` overrides only the global Qwen stage. The profile fixes the global step count at four because its Lightning LoRA is distilled for that schedule; the face stage uses its own eight-step schedule. `--model` is unsupported. `--tile` follows the global-only route described above, with one full-frame face stage after blending.
-
-Direct comparison now covers two official upstream before/after pairs plus the existing crowded `gemini_3` fixture. The published upstream examples were scored against their own original inputs, with the upstream output resized back only for metric alignment where necessary:
-
-| Case | Result | ArcFace identity | Face LPIPS | Texture retention | Image LPIPS | SSIM |
-|---|---:|---:|---:|---:|---:|---:|
-| Upstream example 10 | published upstream | 0.976 | 0.172 | 0.166 | 0.259 | 0.627 |
-| Upstream example 10 | local `qwen-zimage` | 0.950 | 0.045 | 0.570 | 0.167 | 0.765 |
-| Upstream example 10 | current polished ControlNet | 0.701 | 0.105 | 0.941 | 0.094 | 0.781 |
-| Upstream example 12, matched size | published upstream | 0.976 | 0.014 | 0.873 | 0.111 | 0.777 |
-| Upstream example 12, matched size | local `qwen-zimage` | 0.947 | 0.015 | 0.708 | 0.085 | 0.896 |
-| Upstream example 12, matched size | current polished ControlNet | 0.548 | 0.061 | 0.961 | 0.105 | 0.887 |
-
-The result reproduces the upstream architecture's main advantage: identity retention is far stronger than the current ControlNet path. On the group example, local face LPIPS nearly matches the published upstream output and whole-image fidelity is better; upstream still leads slightly on ArcFace identity and texture retention. ControlNet preserves more global detail and, on example 10, lower provisional OCR CER, but its faces drift to different identities. The OCR reference for example 10 came from the original image's OCR rather than hand transcription, so it is supporting evidence, not a text certification. The published upstream outputs are also downscaled relative to their originals, which penalizes their detail metrics but is the actual result the repository presents.
-
-The comparison exposed a real implementation defect on a non-/16 input: the requested DiffSynth dimensions were floored while the PIL image remained at its original size, so the VAE latent and noise grid disagreed. Regression tests were written to fail on that mismatch, then both global and face inputs were changed to use the exact same aligned grid as their `height` and `width`.
-
-**Final candidate oracle result (2026-07-25):** the user checked every image in the provider-separated `full-clean-final-candidate-2026-07-25-by-oracle` bundle with the corresponding provider oracle and confirmed that none of the six outputs retained SynthID or the provider generation signal. These are the current seed-0 bytes after the complete `visible -> qwen-zimage -> metadata` route, including the calibrated YuNet 0.5 gate and the prompt-cache/model-residency optimizations. This supersedes the earlier first-port batch check as the release-candidate result. It certifies these exact outputs, not every seed, resolution, or content class.
-
-YuNet's score threshold is 0.5, not the upstream graph's YOLO threshold of 0.2: detector scores are not interchangeable. The copied 0.2 threshold admitted false/duplicate boxes and multiplied serial Z-Image calls. The calibrated gate retained every visible face in the public and upstream fixtures while reducing `gemini_3` from 36 boxes to 18 and the poster from 30 to 10. Serial face regeneration still scales with the retained detector count. Visual QA also found that the smallest multilingual text degraded on the typography sheet even though the larger headings survived. Keep `controlnet` as the compatibility and cost default, but recommend `qwen-zimage` when the user prioritizes output fidelity, especially face identity. The final exact-output oracle check covers the current YuNet threshold and runtime optimizations; do not call the profile broadly certified until a wider seeded face/text matrix is complete.
-
-**Modal runtime measurement (2026-07-24 through 2026-07-25, seed 0, GPU stage only):** the exact paired A100-40GB run measured ControlNet at 3.342-12.543 seconds per image. `qwen-zimage` took 133.556-188.493 seconds on the three zero-face images and 1212.496 seconds on the 18-face group. The same group initially took 262.072 seconds on an exact H100, including 181.764 seconds in serial face regeneration. On H100 the three zero-face cases took 45.029-65.071 seconds. The shipped fast-load resident placement reduced the group to 133.543 seconds total and 38.272 seconds for face regeneration while producing a pixel-identical output; peak CUDA allocation rose from 24.364 to 43.477 GiB. Setup increased from 32.282 to 43.960 seconds, so even a cold one-request total fell from 294.354 to 177.503 seconds. Reusing the fixed prompt embeddings reduced a warm 18-face request further to 78.474 seconds after an earlier request populated the Qwen embedding; the cached and uncached outputs were pixel-identical, and peak VRAM was unchanged. The Qwen cache helps from the second request in one container, while the Z-Image cache helps after the first face in a multi-face request. Residency is automatic at 64 GiB VRAM or above; smaller cards retain offload. H100 remains both faster and cheaper at the live Modal rates for this workload. Pricing is intentionally not copied here; calculate from the current Modal rate and the recorded GPU seconds. Model setup must be added to an un-warmed single call or amortized over a warm batch.
+See [scope, safety, and legal notes](legal-and-safety.md).

docs/legal-and-safety.md

@@ -0,0 +1,88 @@
+# Scope, safety, and legal notes
+
+This page explains the project's intended boundary. It is not legal advice.
+Laws and platform rules change, so check the current rules that apply to your
+location and use case.
+
+## Intended scope
+
+The project removes AI provenance marks that a platform adds to content the
+user generated or edited themselves. Examples include:
+
+- visible AI generation labels;
+- invisible provenance watermarks;
+- C2PA and metadata based AI disclosures.
+
+The purpose is user control over the user's own output, false positive cleanup,
+interoperability work, and watermark robustness research.
+
+## Out of scope
+
+The project does not provide automatic removal for marks that protect a third
+party's paid or copyrighted asset, including:
+
+- stock agency previews;
+- marketplace and classifieds marks;
+- tiled overlays used to gate a purchase;
+- artist protection systems such as Nightshade or Glaze.
+
+The `erase` command is a generic region tool. Users are responsible for having
+the right to edit the selected content.
+
+## What removal does not prove
+
+Removing a local signal does not:
+
+- prove that an image is human made;
+- remove server side generation history;
+- anonymize the generating account;
+- defeat every statistical AI detector;
+- guarantee that a provider's current verifier will reject the result;
+- make deceptive or unlawful use permissible.
+
+An original file may remain linked to an account or generation session in a
+provider's systems even after a local copy is changed.
+
+## Legal context
+
+Some jurisdictions and platforms require AI generated content to carry visible
+or machine readable disclosures. Rules may apply to providers, publishers,
+users, or a combination of them. Some laws also restrict removing or
+suppressing provenance information.
+
+Before removing a mark, consider:
+
+1. whether you own or are authorized to edit the content;
+2. whether a disclosure is legally required where the content will be used;
+3. whether removal would mislead a viewer about authorship or origin;
+4. whether copyright management information is involved;
+5. whether a platform's terms prohibit the change.
+
+The repository does not provide jurisdiction specific legal advice. The user is
+responsible for checking current law and policy.
+
+## Appropriate uses
+
+Examples that fit the project scope include:
+
+- removing metadata that exposes an account identifier from your own file;
+- correcting an AI label applied to a human photograph after a limited edit;
+- publishing your own generated artwork under the disclosure rules that apply
+  to you;
+- testing the robustness of watermarking and provenance systems;
+- evaluating image processing pipelines in a controlled environment.
+
+## Uses the project does not condone
+
+- fraud or impersonation;
+- nonconsensual sexual imagery;
+- hiding copyright infringement;
+- presenting generated content as human made where that claim is deceptive;
+- evading a disclosure that the law requires;
+- removing protection from someone else's paid asset.
+
+## Reporting security or safety concerns
+
+Open a GitHub issue when the concern can be discussed publicly without exposing
+private data. Do not attach confidential images, credentials, or personal
+information to a public issue.

docs/module-internals.md

@@ -1,435 +1,391 @@
 # Module internals
 
-> Relocated verbatim from `CLAUDE.md` on 2026-06-11 to keep the always-loaded
-> context small. Long single-line entries were reformatted into paragraphs;
-> no content was changed or summarized.
+This page documents the current implementation contract. It intentionally
+avoids experiment logs, corpus counts, and calibration history. Those records
+live in [the verification plan](verification-plan.md) and the research archive
+listed in [the documentation index](index.md).
 
-Full per-module detail: design decisions, tuned thresholds, calibration
-history, incident records, and the regression-guard map. The compact module
-list lives in `CLAUDE.md`; read the relevant section here before changing a
-module.
+Read the relevant section before changing a subsystem. When this page and the
+code disagree, the code and its tests are authoritative and this page must be
+updated in the same change.
 
-## `noai/c2pa.py`
+## Architecture
 
-`noai/c2pa.py` — C2PA reading, **official c2pa-python `Reader` first, hand-rolled parser as fallback** (migrated 2026-06-18; the official lib is a core dep, MIT/Apache, spec-tracking). `read_manifest_store_json(path)` runs `Reader.try_create` with a default `Context` (NO trust enforcement — we report what is in the file, we do not gate on cert trust) and returns the **whole** manifest-store JSON (every manifest plus ingredient manifests); it is memoized per (path, mtime) (`lru_cache(maxsize=8)`) because one `identify`/`get_ai_metadata` call invokes the structured parser ~3x on the same file. `extract_c2pa_info(path)` builds its dict from that store JSON (`_info_from_store_json`: structured `claim_generator` from the active manifest's `claim_generator` / `claim_generator_info[].name`, `timestamp` from `signature_info.time`) and falls back to the legacy caBX parser (`_extract_c2pa_info_png`) when the reader is unavailable (broken/absent wheel, `reader_available()` False) or finds no parseable manifest (synthetic/partial test blobs, the inject round-trip's re-stitched chunk). **Both paths share `_populate_registry_fields(buf, info)`** — the issuer / AI-tool / action / source-type / SynthID / soft-binding registry byte-scan applied to the store JSON (reader path) or the raw caBX bytes (fallback) — so the return-dict shape is identical and the registry stays the single source of truth. Whole-store scanning is load-bearing: a ChatGPT *edit* of a Sora generation keeps `trainedAlgorithmicMedia` + issuer "OpenAI" on the **parent/ingredient** manifest, not the active "opened" one (the active manifest's `signature_info.issuer` is "OpenAI", `common_name` "Truepic Lens CLI in Sora", so the issuer field now reads "OpenAI, Truepic" — first-match-wins platform attribution still resolves OpenAI). `extract_c2pa_info` now also serves non-PNG containers (JPEG/AVIF/MP4) structurally via the reader; the consumers (`identify`, `synthid_source`, `get_ai_metadata`) already merge `info OR byte-scan`, so this strictly upgrades the non-PNG path with no double-counting. `synthid_watermark`/`synthid_vendors` is set when the manifest is signed by a SynthID-using vendor on AI content; `soft_binding`/`soft_binding_vendors` when a `c2pa.soft-binding` `alg` names a forensic-watermark vendor (`soft_binding_vendors_in(buffer)` is the shared byte-scan, used by both paths and the non-PNG binary path). `extract_c2pa_chunk` / `inject_c2pa_chunk` / `has_c2pa_metadata` stay the PNG caBX byte tools (raw-chunk extraction for `extractor.py`, test injection, fallback detection). PNG/caBX chunk reads are clamped to the remaining file size (`safe_length = min(length, remaining)`; skipped chunks use seek) so a malformed huge `length` cannot drive a multi-GB allocation (shared safety discipline matching `isobmff.scan_c2pa_region`). Regression-guarded by `tests/test_noai.py::TestC2PARealSamples::{test_extract_info_uses_reader_store,test_fallback_to_png_parser_when_reader_unavailable}`.
+The package has four main paths:
 
-## `noai/constants.py`
+```mermaid
+flowchart LR
+    Input[Input file] --> Identify[Identify provenance]
+    Input --> Visible[Visible mark removal]
+    Input --> Invisible[Diffusion regeneration]
+    Input --> Metadata[Metadata stripping]
 
-`noai/constants.py` — PNG_SIGNATURE, C2PA_CHUNK_TYPE, C2PA_SIGNATURES, and `C2PA_AI_VENDORS` — the single `C2paAiVendor` registry of C2PA-signing vendors (issuer byte, resolved org name, the `identify` platform label, and a `synthid` flag), from which `C2PA_ISSUERS`, `SYNTHID_C2PA_ISSUERS` (issuers that pair SynthID with C2PA: Google, OpenAI), and `identify._ISSUER_PLATFORM` are all **derived** — plus `C2PA_SOFT_BINDINGS` (soft-binding `alg` prefix → forensic-watermark vendor: Adobe TrustMark, Digimarc, Imatag, Steg.AI, Microsoft, ...). Add a new C2PA vendor as one `C2PA_AI_VENDORS` entry (never edit the derived dicts), a new soft-binding to `C2PA_SOFT_BINDINGS`; not inline. A vendor that signs under multiple legal names needs one entry PER distinctive issuer byte string: e.g. ByteDance's Volcano Engine is registered both as latin `volcengine` AND the Chinese legal entity `北京火山引擎科技有限公司` (UTF-8; the latin needle misses the Chinese-named certs entirely) — both normalize to the same "ByteDance" needle/platform. ElevenLabs ("Eleven Labs Inc.", pure generative-AI) is registered as a generator. A vendor may also set **`asserts_ai=True`** — its presence asserts AI generation even without a `trainedAlgorithmicMedia` digital-source-type; the derived `C2PA_IDENTITY_AI_ORGS` frozenset feeds `identify`, which lifts the AI verdict for such an issuer. Set it ONLY for a pure-generator brand with a distinctive issuer/generator string: **Dreamina** (ByteDance's international Jimeng brand, signed as "Bytedance Pte. Ltd." with a `Dreamina/x.y` claim generator and NO source-type — the caBX / store-JSON byte-scan sees the `Dreamina` token across active + ingredient manifests, where the active one is often a plain `c2pa-tool` transcode; verified on the retained corpus 2026-07; normalizes to the shared "ByteDance" needle/platform). Do NOT set `asserts_ai` on common-word issuers (Adobe/Google/OpenAI/Microsoft) — they appear incidentally in unrelated XMP/trust-chain bytes, so they must stay source-type-gated. Deliberately EXCLUDED (mined-corpus candidates 2026-06-20, documented in the file): TikTok Inc. (a content-provenance / AI-labeling signer on uploads, not a generator) and PixelBin.io / "Fynd" (an image transform / CDN signer) — registering either as a generator would mis-label human uploads as AI; the `is_ai` verdict keys off the digitalSourceType, which is already honored.
+    Identify --> Report[ProvenanceReport]
+    Visible --> VisibleOutput[Localized and filled image]
+    Invisible --> InvisibleOutput[Regenerated image]
+    Metadata --> MetadataOutput[Container with AI metadata removed]
+```
 
-## `metadata.py`
+The `all` command runs visible removal, optional invisible regeneration, and
+metadata stripping in that order.
 
-`metadata.py` — `scan_head(path, size=1MB)` is the shared input for every C2PA/AIGC/IPTC byte scan: first `size` bytes plus the payloads of any provenance metadata found beyond that window — for ISOBMFF, the late provenance boxes from `isobmff.scan_c2pa_region` (catches a manifest after a large `mdat`); for **PNG**, the late `tEXt`/`iTXt`/`zTXt`/`eXIf`/`iCCP` chunks from `_png_late_metadata` (catches an XMP/EXIF packet appended after a large `IDAT`, e.g. a TC260 AIGC label at ~2.7 MB). Behavior-neutral (`f.read(size)`) for non-ISOBMFF inputs and for any file that fits within `size`. Use it instead of `open().read(1MB)` for any new marker scan.
+## Command line interface
 
-**Memoized per (path, size, mtime)** (added 2026-06-09, `_scan_head_cached` lru_cache, `maxsize=8`): one `identify`/`get_ai_metadata` call fans out to ~8 byte-scan detectors that each re-read the same file head, so the cache turns those into a single read; the mtime key invalidates on change, a stat failure falls back to an uncached read. `synthid_source(path)` returns the vendor name(s) if the C2PA manifest implies a SynthID pixel watermark, else None. Format-agnostic: PNG via the caBX parser, JPEG/WebP/AVIF/HEIF/JXL via a binary scan (C2PA marker + SynthID issuer + AI-source marker). `get_ai_metadata` surfaces the verdict, and `metadata --check` prints it as a callout. Both `get_ai_metadata` and `has_ai_metadata` guard the PIL open with `except Exception` (HEIC/unknown formats raise non-OSError) and fall through to the binary scan. `xai_signature(path)` detects xAI/Grok's EXIF-only scheme (`ImageDescription` = `Signature: <base64>` + UUID `Artist`); it feeds `has_ai_metadata`, `get_ai_metadata` (key `xai_signature`), and `identify`. `iptc_ai_system(path)` detects the IPTC Photo Metadata 2025.1 AI-disclosure XMP properties (`IPTC_AI_FIELD_MARKERS` = `AISystemUsed`/`AISystemVersionUsed`/`AIPromptInformation`/`AIPromptWriterName`) and returns the `AISystemUsed` generator name (or `"fields present"`). `remove_ai_metadata` routes **ISOBMFF video** (`.mp4`/`.mov`/`.m4v`) through the same `isobmff.strip_c2pa_boxes` as AVIF/HEIF (MP4 is ISOBMFF), and `_scrub_ai_exif` removes the xAI signature + AI-generator EXIF tags on JPEG output. `strip_c2pa_boxes` is **fail-safe** on a malformed box: it returns the original bytes unchanged with a logged warning instead of truncating the tail to EOF (detection-only `scan_c2pa_region` still stops at a malformed box). `_png_late_metadata` clamps each late-chunk read to the remaining file size (`safe_length = min(length, remaining)`) so a malformed `length` cannot drive a multi-GB allocation, AND advances the cursor by `safe_length` (not the raw `length`) so an inflated length cannot jump past EOF and abort the scan, silently skipping a genuine AI-label chunk after it.
+[`cli.py`](../src/remove_ai_watermarks/cli.py) owns command parsing and
+user-facing exit behavior.
 
-## `identify.py`
+Important contracts:
 
-`identify.py` — the OpenAI rollout caveat is keyed on `_vendor_of(synthid) == "OpenAI"` (not a raw substring over the issuer + verdict blob). `identify(path)` aggregates every locally-readable signal (C2PA issuer→platform, C2PA soft-binding forensic-watermark vendor, **C2PA cloud-manifest reference** via `metadata.c2pa_cloud_manifest` — signal `c2pa_cloud`, **medium**, provenance-only (does NOT set `is_ai`, excluded from `ai_from_metadata` + clash vendors): a C2PA 2.4 Durable-Content-Credentials case where the embedded manifest is stripped but an XMP `dcterms:provenance` pointer to the vendor's cloud manifest store (`_C2PA_MANIFEST_REPOSITORIES`, today `cai-manifests.adobe.com` → "Adobe Content Authenticity") survives, so the credentials stay recoverable server-side; only emitted when no embedded manifest already attributed the file — surfaced on 2 corpus PNGs 2026-06-10 that read fully `unknown` before, IPTC "Made with AI" + IPTC 2025.1 `AISystemUsed`, embedded SD/ComfyUI params, SynthID proxy, xAI/Grok EXIF signature via `metadata.xai_signature`, the China TC260 AIGC label via `metadata.aigc_label`, the HuggingFace `hf-job-id` job marker via `metadata.huggingface_job`, the Samsung Galaxy AI editing marker via `metadata.samsung_genai`, the visible marks — Gemini sparkle plus the registered vendor marks, including Tencent Yuanbao 元宝 / AI生成 and Samsung Galaxy AI "Contenuti generati dall'AI" text marks via the `watermark_registry` — open invisible watermark, Adobe TrustMark via `trustmark_detector`) into one `ProvenanceReport`. `is_ai_generated` is True or None (never asserted False — stripped metadata is not proof of clean origin). The `hf_job`, visible-mark, and Samsung `samsung_genai` signals are **medium** confidence: each lifts an otherwise-Unknown verdict to a tentative AI (`hf_only` / `visible_only` / `samsung_only`, parallel branches; `visible_only` fires on any `visible_*` signal) but is excluded from the high-confidence `ai_from_metadata` set, so none overrides a hard metadata signal.
+- Single-image arguments reject directories.
+- `visible` writes no output when no registered mark is selected and exits with
+  `EXIT_NO_VISIBLE_MARK`.
+- `invisible` writes no output when no supported local signal is found, unless
+  `--force` is supplied.
+- The two no-signal conditions currently share exit code `2`.
+- Hard processing and write failures exit with code `1`.
+- `all` can still write the completed visible and metadata stages when the
+  diffusion dependencies are unavailable, but exits with code `1` so the
+  partial result is not reported as complete.
+- `batch` counts per-file failures and exits nonzero if any file failed or an
+  applicable invisible stage was skipped because its dependencies were absent.
 
-**AI-generated vs AI-enhanced** (`ProvenanceReport.ai_source_kind`, roadmap item): the C2PA digital-source-type is split into `"generated"` (trainedAlgorithmicMedia, fully synthetic) vs `"enhanced"` (compositeWithTrainedAlgorithmicMedia, a real photo with an AI-composited region) — the two byte strings are unambiguous (`compositeWithTrainedAlgorithmicMedia` capitalizes the inner "Trained", so a lowercase `trainedAlgorithmicMedia` match is standalone full generation; full generation wins when both appear). `ai_source_kind` is set only when the AI verdict actually came from the C2PA source type (a non-C2PA AI signal — IPTC/AIGC/local gen/xAI — leaves it None). It lets a caller branch a full-frame scrub (`generated`) from a region-targeted clean that preserves the real photo (`enhanced`; see `noai/tiling.feather_region_composite`). The CLI verdict line reads "AI-generated (fully synthetic)" vs "AI-enhanced (real content with an AI-composited region)".
+The decorators for diffusion options are shared by `invisible`, `all`, and
+`batch`. The runtime help generated by Click is the source of truth for option
+names and defaults.
 
-**Visible-mark detection** (`check_visible`, signals `visible_sparkle` plus `visible_<registry-key>` for every registered vendor mark): the Gemini sparkle keeps its own file-level path (`_visible_sparkle` → `gemini_engine.detect_sparkle_confidence`, promoted only at confidence ≥ `_SPARKLE_THRESHOLD`, which is the SHARED `watermark_registry.GEMINI_SPARKLE_TRUST_CONF` (0.5) — imported, not a private copy, so the provenance detect threshold and the removal `detect_marks` / `_gemini_detect` arbitration gate can never drift (the detect-vs-remove desync from roadmap P0#7; regression-guarded by `tests/test_identify.py::TestSparkleDetectRemoveAlignment`, which composites the real demo sparkle at borderline opacities and asserts identify and `detect_marks` AGREE on either side of the line). Lowering the gate to recover faint sub-0.5 sparkles was evaluated 2026-06-20 and REJECTED: a real Doubao text mark scores ~0.40-0.42 as a gemini match with a HIGHER core-ring brightness margin than a genuine faint sparkle, so neither confidence nor the brightness gate separates them in the [0.35, 0.5) band — lowering trades a rare miss for false-positive removals on clean images. Corpus-tuned to separate Gemini sparkles ≥0.56 from non-sparkle ≤0.49), while all registered vendor marks reuse the registry detectors (`_visible_text_marks` → `watermark_registry`, iterating `_VISIBLE_MARK_PLATFORM`), each gated by its own calibrated engine threshold via `MarkDetection.detected`. Doubao/Jimeng are normally also caught by the TC260 AIGC metadata label and Samsung by its C2PA + `genAIType` marker, so the visible path is their stripped-metadata fallback. Visible marks set `platform` only when no harder signal already did, and (like the sparkle) are excluded from integrity-clash vendor claims. The cv2 dependency lives in the engines, not here.
+The deprecated `--auto` option does not select a pipeline or change adaptive
+polishing. [`_resolve_auto_polish`](../src/remove_ai_watermarks/cli.py) emits a
+warning and returns the explicit polish value unchanged.
 
-**`import identify` is deliberately light** (~26 MB; ~36 MB with cv2 loaded by a visible-mark run, ~106 MB for a full `check_visible` run): it imports the `noai.c2pa`/`noai.constants` submodules, and `noai/__init__` is lazy (see "Test and lint"), so torch/diffusers are NOT pulled at import even in a full `gpu`/`detect` install — fits a 512 MB host. `noai.c2pa` does eagerly import the **c2pa-python** binary (Rust + cryptography, ~+5 MB RSS, no torch) for the primary `Reader` path — light enough to stay on the dependency-light host; a broken/absent wheel degrades to the byte-scan parser (`reader_available()` False). The heavy paths are opt-in: `check_invisible=True` needs the `detect`/`trustmark` extras (each pulls **torch**; TrustMark also **downloads weights**), so on a core-only deploy leave `check_invisible` off (it is a no-op there anyway). Before the lazy `__init__`, the mere presence of torch in the env inflated `import identify` to ~420 MB.
+Regression coverage:
 
-**C2PA platform attribution is device-token-first, issuer-scan fallback** (`_device_platform` scans manifest bytes for `_DEVICE_C2PA_PLATFORM` tokens, then `_attribute_platform`/`_ISSUER_PLATFORM`).
+- [`test_cli.py`](../tests/test_cli.py)
+- [`test_cli_robustness.py`](../tests/test_cli_robustness.py)
+- [`test_optional_deps.py`](../tests/test_optional_deps.py)
 
-**Why, verified on real signed files 2026-05-26:** the old issuer-only byte-scan matched ANY issuer substring anywhere, so multi-entity manifests mis-attributed -- Leica→"Truepic" (a signing authority in the trust chain), Nikon→"Adobe Firefly" (XMP-toolkit "Adobe" + the sample's "Adobe_MAX" name), Pixel→"Google (Gemini)" ("Google LLC" cert org), Truepic→"Google". A distinctive device token wins instead.
+## High-level Python API
 
-**Token distinctiveness is load-bearing:** bare `b"Truepic"` mis-fires (it appears in unrelated trust chains -- it mis-attributed the OpenAI `chatgpt-1.png` fixture), so the token is the specific `b"Truepic_Lens"` from the Lens SDK claim generator; likewise `b"Pixel Camera"` (cert CN) not bare `b"Pixel"`. `_DEVICE_C2PA_PLATFORM` lists ONLY tokens **verified against a real C2PA file**: Leica (`lc_c2pa`/`Leica Camera`), Nikon (`NIKON`), Pixel (`Pixel Camera` -- from a real Pixel 10 Pro file attached to c2pa-rs issue #1609/#1554), Sony (`sony.sig`/`sony.cert` -- Sony's own C2PA assertion namespace, verified on a real Sony PXW-Z300 file; NOT bare "Sony" which is a common EXIF Make), Truepic (`Truepic_Lens`). Canon/Bria have **no public direct-download C2PA sample** (checked exhaustively: GitHub issue/PR attachments, contentcredentials gallery, HF datasets -- all upload-to-verify or token-gated; Canon's only public file was a self-signed hobbyist CR3, not factory), so they stay unmapped until a real file is captured (same fixture discipline as Grok/Doubao). The Sony sample is video (MP4) -- our ISOBMFF C2PA path detects it; Sony Alpha stills likely share the `sony.*` namespace but are not separately verified.
+[`api.py`](../src/remove_ai_watermarks/api.py) provides:
 
-**Samsung Galaxy + ASUS Gallery live in a separate `_SIGNER_C2PA_PLATFORM` (scanned after `_device_platform`, before the issuer fallback), NOT in `_DEVICE_C2PA_PLATFORM`** — verified on real signed files 2026-05-29. Reason: a Galaxy phone stamps BOTH its device cert AND a `trainedAlgorithmicMedia`/genAIType AI marker on a Generative-Edit image, so treating it as a "genuine camera capture" would false-fire integrity-clash rule 2 on every Galaxy AI edit. The signer tokens (`b"Samsung Galaxy"` cert org — distinct from the EXIF `SM-xxxx` model string on ordinary Samsung photos; `b"com.asus.gallery"` claim generator) only resolve the platform label; the AI verdict still comes from the source-type / genAIType. ASUS Gallery is a C2PA-signed edit with no AI marker, so it attributes the platform without asserting `is_ai`.
+- `remove_visible`
+- `visible_provenance`
 
-**Samsung's `genAIType` (in the proprietary `PhotoEditor_Re_Edit_Data` JSON) is an undocumented Galaxy-AI editing marker** (`metadata.samsung_genai`, gated on the `PhotoEditor_Re_Edit_Data` container; non-zero value = AI tool used, values {1,5} observed; Galaxy AI appends it as a trailer AFTER the JPEG EOI, so `samsung_genai` reads the file TAIL when the 512 KB quick-scan head misses it — else a multi-MB photo's trailer past the window went undetected while removal, which reads the whole file, would still strip it; removal truncates the post-EOI Samsung trailer via `metadata._strip_samsung_trailer`, pixels bit-identical): medium-confidence because the field has no public spec (verified 2026-05-29: absent from C2PA spec + Samsung docs), but it co-occurred with `trainedAlgorithmicMedia` in 3/3 verified files that record a source-type and was the SOLE AI marker on a Galaxy S24 file that omits the source type. Camera C2PA marks capture authenticity, not AI (Pixel carries `computationalCapture`, not `trainedAlgorithmicMedia`), so these never set `is_ai` -- that stays driven by digital-source-type. `c2pa.cbor_text_after` (now public) is best-effort for the `generator` detail string only and can be None when the manifest keys it `claim_generator_info` (Pixel).
+The package root exposes both lazily through
+[`__getattr__`](../src/remove_ai_watermarks/__init__.py), keeping a plain package
+import free of the heavier image and model imports.
 
-**Issuer→generator mapping is `is_ai`-gated** (`_attribute_platform(issuers, is_ai=c2pa_is_ai)`): a specific AI-generator platform is named only when the digital-source-type is `trainedAlgorithmicMedia`; on a non-AI source an issuer substring is treated as incidental (an "Adobe XMP" toolkit string in an *unmapped* Canon/Sony capture would otherwise mislabel it "Adobe Firefly"), so it degrades to the neutral "C2PA signer: X" label. **The one exception is an identity-AI issuer** (`c2pa_is_ai = c2pa_source_kind is not None or c2pa_identity_ai`, where `c2pa_identity_ai` is any resolved issuer org in `C2PA_IDENTITY_AI_ORGS`): a vendor flagged `asserts_ai` (today only Dreamina) sets `c2pa_is_ai` True on its own, so its platform resolves even though the manifest carries no `trainedAlgorithmicMedia`. This is safe precisely because the flag is restricted to distinctive brand strings, not the incidental-mention-prone common words. Real Firefly/OpenAI/Google output carries the AI source-type, so it is unaffected (verified: chatgpt-1.png→OpenAI, firefly-1.png→Adobe Firefly still attribute). `_attribute_platform` defaults `is_ai=True` so the mapping stays unit-testable in isolation. Add capture-camera tokens to `_DEVICE_C2PA_PLATFORM`, editing-app/AI-device signer tokens to `_SIGNER_C2PA_PLATFORM`, generator/issuer platforms to the `C2PA_AI_VENDORS` registry in `constants.py` (which derives `_ISSUER_PLATFORM`), not inline. For non-PNG containers (JPEG/WebP/AVIF/HEIF/JXL) the caBX parser returns nothing, so issuer (`_issuers_in`) and generator (`_ai_tools_in`, reusing `C2PA_AI_TOOLS`) are recovered by binary-scanning the first MB. EXIF `Software` / `Make` / `Artist` / `ImageDescription`, XMP `CreatorTool`, and PNG `tEXt` chunks (`Software`/`Source`/`Title`/`Description` — NovelAI stamps its generator there, not EXIF) are read by `metadata.exif_generator` (PIL+piexif for any format PIL opens incl. AVIF, plus a container-agnostic XMP raw-byte scan that also covers HEIF/JXL), matched against `AI_GENERATOR_TOKENS` so ordinary editors (plain "Adobe Photoshop") and real-camera `Make` ("Apple"/"Canon") are not flagged. Tokens mined from the retained corpus 2026-06-22: `novelai`, `reve.com` (full token, not bare `reve`), `aphrodite ai` — all no-C2PA generator stamps that previously read as no-signal (and under the P0#5 no-signal skip would have skipped the scrub).
+For path inputs, `remove_visible` reads provenance metadata, preserves alpha,
+and optionally writes and strips metadata. Array inputs are treated as BGR
+arrays and have no file provenance or separate alpha plane.
 
-**Ideogram tags its output with EXIF `Make="Ideogram AI"`** (verified on a real download 2026-05-24) — that's why `Make` is read.
+When no visible mark is removed, a same-format path copy preserves the original
+bytes. `write_noop=False` leaves the requested output path untouched instead.
 
-**Integrity-clash detection** (`_integrity_clashes`, surfaced as `ProvenanceReport.integrity_clashes`, printed in red by `identify` and serialized to `--json`): contradictions between independent generator stamps are a laundering/spoofing tell. Two rules: (1) two or more distinct AI-origin vendors named by **independent** signals (e.g. C2PA OpenAI + EXIF `Make="Ideogram AI"`), and (2) a camera-capture C2PA device (`_DEVICE_C2PA_PLATFORM`) coexisting with an AI-generation marker **from a source INDEPENDENT of the camera's own manifest**.
+Regression coverage:
 
-**Rule 2's independence gate (added 2026-06-11):** a device that both captures and runs on-device generative AI (Google Pixel Magic Editor / Pixel Studio) records the capture AND the AI edit in ONE C2PA manifest — so the AI vendor is named only from that same manifest (`c2pa` issuer + `synthid` proxy, both `c2pa_manifest` source) — a legitimate edit chain, NOT a clash. Rule 2 therefore fires only when some `ai_vendor_claims` family has a source `!= "c2pa_manifest"` (EXIF/XMP generator, IPTC, TC260 AIGC, a second manifest naming AI on a camera capture — the real laundering tell). This killed a false-positive class on the corpus: 2 real Pixel generative-edit PNGs (`computationalCapture` + `trainedAlgorithmicMedia` + "Applied imperceptible SynthID watermark" in one Google manifest) read as camera-vs-AI clashes before the gate. Pure cameras (Leica/Sony/Nikon/Truepic) that do NOT generate AI still clash on any within-manifest AI marker only if it is independent — they never legitimately carry one, so the gate is behavior-neutral for them while fixing Pixel (regression-guarded by `test_identify.py::TestIntegrityClashesHelper::{test_pixel_generative_edit_same_manifest_no_clash,test_camera_plus_independent_ai_marker_still_clashes}` + `TestIntegrityClashEndToEnd::test_pixel_generative_edit_no_clash`).
+- [`test_api.py`](../tests/test_api.py)
+- [`test_image_io.py`](../tests/test_image_io.py)
 
-**Independence is source-grouped (`_CLASH_SOURCE`, added 2026-06-02):** the C2PA issuer attribution (`c2pa`) and the SynthID proxy (`synthid`) are NOT independent — the proxy is inferred from the *same* manifest — so they share one source and two vendors named within a single manifest do not clash. This killed a false-positive class found on the spaces corpus: legitimate multi-actor manifests where a product wraps another vendor's engine (Microsoft Designer on OpenAI → `OpenAI, Microsoft`; Microsoft on Google → `Microsoft, Google LLC, Google C2PA Core Generator Library`) or an edit chain re-signs (Adobe over a Gemini original → Adobe c2pa + Google synthid) — 19 such files across the 2026-06-01/02 batches read as clashes before the fix. Rule 1 still fires when a manifest vendor disagrees with a genuinely independent stamp (EXIF/XMP generator, IPTC `AISystemUsed`, AIGC, xAI); each non-`c2pa`/`synthid` family is its own source (`test_identify.py::TestIntegrityClashes::{test_multi_actor_manifest_no_clash,test_manifest_vendor_vs_independent_signal_clashes}`). Vendor normalization is `_vendor_of` over `_AI_VENDOR_TOKENS` (so a C2PA "Google (Gemini)" issuer and a SynthID-Google proxy agree, while different vendors clash). `_AI_VENDOR_TOKENS` covers ByteDance (all brands: bytedance/doubao/jimeng/dreamina/volcengine), Canva, ElevenLabs, and Black Forest Labs in addition to the OpenAI/Google/Adobe/... set — without them a transplanted ByteDance/Canva/BFL C2PA manifest next to an independent conflicting stamp was silently missed. **The generic `China AIGC (TC260)` label names no SPECIFIC vendor** (any Chinese generator applies it), so it cannot vendor-conflict in the spoofing sense: when a Chinese TC260-applying vendor (`_TC260_VENDORS`, today `{ByteDance}`) is co-attributed, Rule 1 attributes the label to that vendor (a legit Doubao image carries BOTH a ByteDance C2PA manifest and its own TC260 label and must not clash); against a NON-TC260 vendor (OpenAI etc.) the label stays generic and still clashes as a laundering tell (`test_bytedance_c2pa_plus_own_aigc_no_clash`, `test_foreign_vendor_plus_aigc_still_clashes`, `test_bytedance_c2pa_plus_foreign_generator_clashes`). Corpus-validated: 0 new clashes on 5000 ByteDance/AIGC/Canva/FLUX carriers.
+## Metadata and provenance
 
-**High-precision by design:** only hard generator stamps feed it (C2PA-issuer when source is AI, SynthID, EXIF/XMP generator, IPTC `AISystemUsed`, xAI, AIGC); the fuzzy visible sparkle and the open invisible watermark are **excluded** (both are low-precision/positive-only signals; the open watermark was also historically a by-product of our own SDXL removal pass, until `watermark_remover` was fixed to load the SDXL pipelines with `add_watermarker=False` — it stays excluded as a fuzzy signal regardless). The c2pa vendor is classified from the issuer attribution / generator, NOT the resolved `platform` (a camera label like "Google Pixel" would mis-normalize to "Google"). All real single-origin fixtures (chatgpt/firefly/doubao/grok/mj) verified to produce **zero** clashes (false-positive guard in `test_identify.py::TestRealSamplesHaveNoClash`).
+### C2PA
 
-**`ai_from_metadata` field + `has_invisible_target` helper (P0#5, 2026-06-22):** the high-confidence union (everything that sets `confidence == "high"`: C2PA AI-issuer / SynthID proxy, IPTC, AIGC, local gen params, EXIF/xAI, open DWT-DCT / TrustMark — the medium-confidence `hf_only`/`visible_only`/`samsung_only` are excluded) is now surfaced as the public `ProvenanceReport.ai_from_metadata` boolean, so callers gate on intent rather than on the `confidence` string. `has_invisible_target(path)` wraps `identify(path, check_visible=False, check_invisible=True)` and returns that field — it is the decision gate for the diffusion scrub (the CLI `invisible`/`all`/`batch` no-signal skip, `cli._no_invisible_signal_exit`): a visible-only or no-signal image has it False, so regeneration (which would only degrade a clean image) does not run. It fails SAFE — any detector exception returns True so the removal still runs (leaving a watermark on a paid removal is worse than over-regenerating). It does NOT prove a pixel SynthID is absent (SynthID is detectable only via its metadata proxy, gone once stripped), so a False means "no locally-detectable target", never "clean". Guarded by `test_identify.py::{TestIdentifyRealSamples::test_has_invisible_target_*,TestHasInvisibleTargetFailSafe}`.
+[`noai/c2pa.py`](../src/remove_ai_watermarks/noai/c2pa.py) reads C2PA with the
+official `c2pa-python` reader first. Its byte-level PNG parser remains a fallback
+for partial and synthetic fixtures that the official reader rejects.
 
-## `watermark_registry.py`
+Vendor attribution comes from the registry in
+[`noai/constants.py`](../src/remove_ai_watermarks/noai/constants.py). Derived
+issuer and platform maps should not be maintained separately.
 
-`watermark_registry.py` — **single catalog of known visible watermarks**, the unified "find known marks in their usual places, recognize, remove" entry.
+### Metadata scanning and stripping
 
-**Localize -> fill by policy (replaced reverse-alpha):** each mark is localized to a binary full-frame footprint mask (a `Localization`), and one shared, swappable fill inpaints that mask via `fill(image, mask, backend=...)` (delegates to `region_eraser.erase`). This replaced the old reverse-alpha removal (invert a captured alpha map, `original = (wm - a*logo)/(1-a)`, plus a thin residual inpaint) for ALL marks — all registered marks, including Yuanbao. **Why it changed:** reverse-alpha depended on a fixed captured alpha map at a fixed position, so it broke whenever a vendor moved or re-rendered its mark; and it was not color-lossless even with the right map (it amplifies 8-bit quantization and JPEG-chroma error by `1/(1-a)`), which showed up as "the color just changed, not removed" reports. Localize -> fill has a benign failure mode: a slightly-off localization just inpaints a small region near-losslessly instead of leaving a color-shifted smear. The captured alpha maps are still used to DETECT the marks and to shape the mask (gemini's footprint), but NOT for pixel recovery. Fill backends: `cv2` (classical inpaint, no deps, the floor), `migan` (MI-GAN ONNX, light, the memory-tight pick where LaMa will not fit), `lama` (big-LaMa ONNX, best quality, heavier, auto-preferred when a learned backend is available); `auto` = LaMa > MI-GAN > cv2, best available. Each `KnownMark` ties a key to {usual `location`, `in_auto` flag, a `_detect` callable → uniform `MarkDetection`, a `_mask` callable → full-frame footprint mask}; `KnownMark.remove(image, *, backend="auto", provenance=False, force=False)`. Entries today include `gemini` (bottom-right sparkle), `doubao` (bottom-right "豆包AI生成"), `jimeng` (bottom-right "★ 即梦AI"), `qwen` (bottom-right "千问AI生成", Alibaba Tongyi Qianwen), `yuanbao` (bottom-right two-line "元宝 / AI生成", Tencent), `samsung` (bottom-**LEFT** "✦ Contenuti generati dall'AI", Samsung Galaxy AI, Italian locale), and the capture-less `jimeng_pill` (top-left "AI生成"). `detect_marks(image, *, provenance=frozenset())` scans all (strict, for the identify verdict); `remove_auto_marks(image, *, sensitivity="auto", provenance=frozenset(), backend="auto")` removes every detected mark in one pass. **Sensitivity (`auto`/`strict`/`assume_ai`)** decides how hard a borderline mark is trusted: the visual detectors are pixel-based (no metadata needed) and the recall gain comes from relaxing the false-positive gate, not from metadata. `resolve_trust` turns the policy + evidence into the per-mark trust level the engines consume as `provenance = level != "strict"` — `strict` never relaxes; `auto` relaxes only on same-product evidence (metadata provenance for that vendor, or a confidently strict-detected sibling of the same `_PRODUCT_OF` — Doubao and Jimeng are both bottom-right ByteDance but distinct products, so they do NOT cross-relax); `assume_ai` relaxes every mark (the caller asserts AI, e.g. a metadata-stripped screenshot). **Three levels, not two: `strict` / `assumed` / `confirmed`.** Relaxing bypasses the engine's false-positive gate outright, and that bypass is contracted to mean the vendor is CONFIRMED (`GeminiEngine.detect_watermark`'s `trust_provenance`: "external metadata already proves this is a Google generation"). An `assume_ai` caller asserts the image is AI, which says nothing about WHICH vendor, so a mark relaxed on assumption alone must also clear `_ASSUMED_CONF_FLOOR` (`assumed_floor_ok`; gemini 0.50) — see "Assumed-trust confidence floor" below. **Perception / decision / action are separated three ways** (the removal path only; `identify` keeps calling `KnownMark.detect` directly, so its verdict is untouched): `_build_candidates(image)` is PERCEPTION — it runs each detector at both trust levels and packages raw verdicts + the pill's flatness feature into `Candidate`s, no policy; `decide(candidates, Context(sensitivity, provenance)) -> [Decision]` is the pure DECISION arbiter — all keep/drop policy (`resolve_trust` cross-mark corroboration + the assumed-trust floor + the pill gate) in one image-free, unit-testable function (`tests/test_watermark_registry.py::TestArbiter`); then `remove_auto_marks` does the ACTION, localizing -> filling each winner. The Gemini FP gate deliberately stays inside `gemini_engine` (not the arbiter) because `identify` reads that same gated confidence — pulling it out would drift the provenance verdict. Behavior was byte-identical to the pre-arbiter two-pass when the arbiter landed; `assume_ai` has since gained the assumed-trust confidence floor (see below), which deliberately changes its verdict on weak gate-bypassed matches.
+[`metadata.py`](../src/remove_ai_watermarks/metadata.py) contains the shared
+metadata scanners and `remove_ai_metadata`.
 
-**Head-to-head validation (v0.12.1 reverse-alpha vs the current localize -> fill):** run over the full labelled visible-mark set, with the cv2 / MI-GAN / LaMa fills each compared against the old reverse-alpha. **doubao and jimeng are identical** across every backend -- 100% coverage and 100% clearance either way. **gemini** strict coverage is a few points below reverse-alpha's (the deliberate false-positive tightening), but the metadata-stripped faint ones are now mostly recovered by the DEFAULT white-core rescue in the FP gate (`gemini_engine`: a bright near-WHITE core distinguishes a real faint sparkle from a colored bright corner -- ~14/20 recovered at ~1.25% clean false-fire; a learned classifier on the same features measured worse, 2026-07 tier-1), the residual under `assume_ai`; clearance is equal (~98% both), and neither version touches pixels outside the mark box (outside-box PSNR ~99). **Clearance is fill-independent** -- cv2, MI-GAN and LaMa all strip the mark's shape equally, so the re-detect metric does not separate them; the difference is purely the *visual fill quality* on the recovered region, and it is background-dependent. reverse-alpha recovered textured and especially regular/structured backgrounds (a lattice, a grid) more cleanly than any inpaint; **LaMa closes most of that gap** (the best learned backend), **MI-GAN can ghost or hallucinate structure**, and **cv2 smears** (the last-resort floor). This is why `auto` resolves `LaMa > MI-GAN > cv2` (`preferred_inpaint_backend`) and warns once on the cv2 fallback; on flat backgrounds every backend is clean.
+Key contracts:
 
-**`assume_ai` removed (2026-07-19).** It relaxed EVERY mark's false-positive gate on the caller's bare assertion that an image is AI. That assertion says nothing about WHICH vendor or WHERE the mark is, which is exactly what a gate bypass is contracted to require -- before it carried a confidence floor it filled a phantom sparkle on 59.8% of genuine camera photos, and even with the floor it was a statistical gamble rather than an instruction. It also had no place in the product's model: detector finds a mark -> remove it; detector finds nothing -> leave the image alone; the USER sees a mark and says so -> act on that.
+- `scan_head` is the shared cached input for bounded byte scans.
+- JPEG stripping walks metadata segments and preserves the entropy-coded image
+  scan.
+- ISOBMFF containers use
+  [`noai/isobmff.py`](../src/remove_ai_watermarks/noai/isobmff.py).
+- Supported non-ISOBMFF audio and video containers use ffmpeg stream copying.
+- The low-level remover is fail-safe and can copy an undecodable file through
+  unchanged.
+- A caller that reports success must use `strip_and_verify`, which scans the
+  written output for surviving markers.
 
-Removing it collapsed the trust ladder from three levels to two (`strict` / `confirmed`) and took `_ASSUMED_CONF_FLOOR`, `assumed_floor_ok` and the `assumed` level with it. `_keep_pill` lost its `sensitivity` parameter (its assume-arm is gone; the metadata arm and its flatness guard are unchanged). Verified on the 240-image unbiased recall sample: doubao 92%/99%, gemini 96%/80%, jimeng 71%/71%, pill 50%/60% -- identical before and after, so nothing on the default path moved.
+Detection and removal must stay in parity. A new marker is incomplete until the
+scanner can find it, the remover can reach every supported placement, and a
+test proves that it no longer appears in the output.
 
-**The replacement advice is PER MARK, because the forced paths are not equally reliable** (measured 2026-07-19):
+Regression coverage:
 
-| path | reliability |
-|---|---|
-| `erase --region x,y,w,h` | sound by construction -- the user supplies the coordinates |
-| `--mark <text-mark> --no-detect` | reasonable: the forced mask is the real glyph blob, non-empty on 13/13 missed doubao marks |
-| `--mark gemini --no-detect` | **NOT recommended** -- falls back to a fixed default sparkle slot, which covered the true sparkle on only **31% of 97** sparkles the strict gate missed (median offset 63px up-and-left). The other 69% fill a clean corner AND report a removal that did not happen. |
+- [`test_metadata.py`](../tests/test_metadata.py)
+- [`test_noai.py`](../tests/test_noai.py)
+- [`test_security_clamp.py`](../tests/test_security_clamp.py)
 
-`cli._no_visible_mark_exit` therefore recommends `erase --region` first and a named text mark second, and never suggests forcing gemini. It previously recommended `--sensitivity assume-ai`, i.e. the product's own hint contradicted its model.
+### Provenance report
 
-**Migration is LOUD, not silent.** `Sensitivity` is a `Literal` and unenforced at runtime, so a 0.15 caller passing `sensitivity="assume_ai"` would otherwise get `auto` behaviour in silence -- a quiet semantic change on exactly the release where they need telling. `validate_sensitivity` (called by `api.remove_visible` and by `Context.__post_init__`) raises a `ValueError` naming the replacement. Regression: `tests/test_watermark_registry.py::TestNoBlanketRelaxation`.
+[`identify.py`](../src/remove_ai_watermarks/identify.py) combines metadata,
+registered visible marks, and optional open invisible-watermark decoders into a
+`ProvenanceReport`.
 
-**Continuous top-hat detection front-end (`detect_frontend` / `tophat_response`, 2026-07-18).** `extract_mask` thresholds the white top-hat into a 0/255 glyph blob and correlates a binary silhouette against it. That is fine for a mark stamped bold and opaque, and destructive for a faint one: a thin translucent overlay shatters into specks under the threshold, and no template can match a blob that is not there (千问 measured 0.170 mean NCC, **0%** over its gate, against doubao's 0.723 / 82% -- same pipeline, each with its own template). The `tophat` front-end never binarizes: the saturation and absolute-luma gates become WEIGHTS instead of hard cuts, so a faint stroke contributes in proportion to its strength, and the response is max-normalized, which makes the score contrast-invariant.
+`is_ai_generated` is `True` or `None`; absence of evidence is not reported as a
+human-made verdict. `ai_source_kind` distinguishes fully generated content from
+AI-enhanced composites when the source metadata provides that distinction.
 
-Doubao is switched to it; jimeng and samsung stay `binary` until measured, because a front-end change must be measured per mark before it ships. Corpus effect on the 240-image unbiased recall sample:
+TrustMark is reported as a watermark signal but does not by itself assert AI
+origin because it can also protect human-authored content.
 
-| mark | recall before | recall after | precision before | precision after |
-|---|---|---|---|---|
-| doubao | 89% | **92%** | 99% | **99%** |
-| jimeng / gemini / pill | unchanged | unchanged | unchanged | unchanged |
+Regression coverage:
 
-**The gate is FRONT-END SPECIFIC and must be re-calibrated, not ported.** The continuous response scores higher overall (mean 0.809 vs 0.723 on the same 90 positives), so the binary-era 0.40 left the provenance-relaxed gate (x0.7) far too low: at 0.40 the arm ran 96% recall / 91% precision (8 false fires), at 0.50 it runs 92% / 99% (1 false fire). 0.50 was chosen because it beats the binary front-end on recall at IDENTICAL precision -- a front-end that only trades one for the other would not have been worth shipping. A first pass at 0.40 also silently depressed the PILL (recall 50% -> 33%), because `_keep_pill` suppresses the pill whenever doubao fires; a coupling worth remembering when tuning any bottom-right mark.
+- [`test_identify.py`](../tests/test_identify.py)
+- [`test_trustmark_detector.py`](../tests/test_trustmark_detector.py)
+- [`test_invisible_watermark.py`](../tests/test_invisible_watermark.py)
 
-**The removal MASK must ride the same front-end, and how it does so was fixed twice.** `tophat` detection does not binarize, but `extract_mask` (which bounds the fill) still does, so a mark faint enough to be found only by the continuous response produced an EMPTY binary blob: `localize` returned `mask=None`, `remove()` was a silent no-op, and `identify` reported `visible_doubao` while `visible` said "no visible mark" on the same file (corpus-measured 2026-07-20: 57 of 60 sampled still-detected Doubao marks untouched, ~8% of its detections). The FIRST fallback (2026-07-19) thresholded the continuous response and took the bounding box of everything above the level -- but the level was `0.5` compared against the max-normalized **uint8 0..255** response, so it selected every non-zero pixel and filled ~120% of the corner box on textured frames (a padded whole-ROI box). It passed parity (a mask that fills everything is trivially detector-clean) and its regression test (a FLAT fixture, where the response is non-zero only on the glyph, so every threshold yields the same box). The SECOND fallback (2026-07-20) uses the detector's OWN best-match box instead: `_tophat_score` was split into **`_tophat_best(image, loc) -> (score, box)`**, the single method whose score gates detection and whose argmax box bounds the mask -- so the two cannot drift by construction, which is how the mismatch arose in the first place. Measured over 14 real faint-path frames (cv2 fill, detector re-run after): the match box fills a **58.7%**-median corner box vs the threshold's **120.9%**, both 100% detector-clean. The largest-connected-component alternative was tighter (10.5%) but removed the mark on only 21% of frames, so it does not cover it and was rejected. Regression: `tests/test_text_mark_faint_mask.py`, whose fixture now carries texture (the flatness of the old one is exactly why it could not see the threshold bug -- mutating the constant to 99.0 left it green). **Any future front-end change must move both the detection and the mask path, or re-check this.**
+## Visible mark removal
 
-**Vendor ATTRIBUTION for the shared-suffix marks -- solved for 千问 by exact sizing, not by a generic template.** The 2026-07-18 worry was that "千问AI生成" and "豆包AI生成" share the `AI生成` tail (four of six glyph cells), so their templates would cross-fire. That held while Qwen was scored at the WRONG SIZE (AUC 0.41-0.59, a coin flip). Re-measured 2026-07-21 at the fitted geometry on real pools (`scripts/vendor_mark_calibrate.py --crossfire`): Qwen's template scores p50 0.224 / p90 0.242 on 400 Doubao-marked frames against a 0.45 gate -- **0 cross-fires**, and 0 on 298 Jimeng-marked frames and 286 clean frames -- because an exact-size 6-glyph template is specific enough that the 2-glyph prefix mismatch dominates the shared tail. Qwen was therefore registered WITHOUT a rival margin (a 0.10 margin would have suppressed ~10% of genuine Qwen detections, whose margin p10 sits at 0.00). A GENERIC shared-tail template remains a harvesting aid, not a detector (measured: 0.407 on a bold positive vs clean p99 0.298).
+### Registry and decision flow
 
-**千问 is registered since 2026-07-21; 星绘 is NOT (still one confirmed example).** The 2026-07-18 measurement below is why 千问 was originally held out -- it is kept because it records the failure CLASS that the registration then had to solve. Measured back then on 14 hand-verified corpus positives, same pipeline, each mark scored with its OWN template:
+[`watermark_registry.py`](../src/remove_ai_watermarks/watermark_registry.py) is
+the only visible-mark registry. `mark_keys()` supplies the CLI choices, so the
+CLI must not maintain a separate mark list.
 
-| mark | n | mean NCC | median | above the 0.40 gate |
-|---|---|---|---|---|
-| doubao | 40 | 0.723 | 0.835 | **82%** |
-| qwen | 14 | 0.170 | 0.179 | **0%** |
+Automatic removal has three distinct stages:
 
-Three candidate explanations were ruled out in order, each by measurement:
-1. **Not the synthetic render.** A template cut from an ACTUAL Qwen mark scores the same as the font-rendered one (real-vs-real 0.307 vs synthetic 0.308), and real masks do not match EACH OTHER.
-2. **Not the morphology kernel.** `MORPH_OPEN`/`MORPH_CLOSE` use fixed 5px kernels regardless of mark size (~9% of a 57px-tall box, ~2.7% of a 188px one). Scaling them with the box height gained +0.014 mean and moved nothing across the gate.
-3. **Not the appearance thresholds.** Sweeping `tophat_delta` / `logo_min_luma` / kernel size peaked at mean 0.35 with 4/14 over the gate.
+1. Perception: each registered detector produces strict and relaxed candidates.
+2. Decision: the pure `decide` arbiter applies sensitivity and corroborating
+   provenance.
+3. Action: each selected mark is localized to a mask and passed to the shared
+   fill function.
 
-The blocker was named as SEGMENTATION on a faint mark -- and the `tophat` front-end (built for Doubao the same week) removed exactly that blocker, yet 千问 still did not register, because the real residual was never segmentation alone: it was **mis-sized geometry** (two size modes the shared 3-rung ladder cannot straddle, plus a locate box that clipped the first glyph). The 2026-07-21 cohort harvest (117 labelled frames) and the full calibration chain are in the `qwen_engine.py` section below and `docs/verification-plan.md`. 星绘 remains where it was: ONE confirmed corpus example, so nothing to calibrate a gate against -- do not register it off a single frame. The synthetic renderer and the evidence chain are kept in `scripts/render_vendor_silhouettes.py`; researched vendor specs are in `docs/watermarking-landscape.md`.
+`sensitivity="strict"` never relaxes a detector. `sensitivity="auto"` can relax
+one only when metadata or a sufficiently strong same-product sibling confirms
+that product. The removed blanket `assume_ai` mode is rejected explicitly.
 
-**RECALL, measured at last (unbiased random sample, 2026-07-18).** Every earlier round sampled where detectors FIRED, so recall was structurally unmeasurable. This round draws 240 images at RANDOM within each provenance class (160 TC260, 80 Google-C2PA) and labels them EXHAUSTIVELY -- both corners shown at native scale, so a missed mark is visible as a miss rather than absent from the data. Build it with `scripts/visible_recall_sample.py`; labels live in the gitignored research dir.
+The Jimeng pill has an additional decision gate because its visual detector is
+weaker than the other registered marks. Keep that policy in the registry, not
+inside unrelated detector engines.
 
-| mark | present | recall | 95% CI | precision | 95% CI |
-|---|---|---|---|---|---|
-| doubao | 90 | **89%** | 81-94% | **99%** | 93-100% |
-| gemini | 46 | **96%** | 85-99% | **80%** | 68-88% |
-| jimeng | 14 | 71% | 45-88% | 71% | 45-88% |
-| jimeng_pill | 6 | 50% | 19-81% | 60% | 23-88% |
+`remove_auto_marks` removes every selected mark, not only the strongest one.
+This matters for images that carry marks in more than one corner.
 
-Effect of the `scale_basis` fix on the same sample (strict verdicts recorded before it): **doubao recall 71% -> 89%**, gemini 91% -> 96%, jimeng 64% -> 71%. Part of the doubao gain is the provenance relaxation rather than the basis alone, since the after-numbers run the full `auto` path.
+Regression coverage:
 
-**Three corrections this forced to earlier numbers:**
-* **Gemini precision is 80% on an unbiased sample, not the 41% the addition-sampled harness reports.** The 41% is precision restricted to relaxation ADDITIONS, which are by construction the marginal cases; production sees mostly strict fires, which are near-perfect. Quote 80% for the product and 41% only when discussing the relaxation arm.
-* Doubao is in excellent shape (89/99) and is no longer the problem it looked like before the basis fix.
-* Landscape is improved but NOT solved: doubao recall by aspect is portrait 92% / square 92% / **landscape 79%**, so a residual geometry gap remains beyond the basis.
+- [`test_watermark_registry.py`](../tests/test_watermark_registry.py)
+- [`test_api.py`](../tests/test_api.py)
 
-**Where the remaining loss actually is:** jimeng and the pill, both at small n with intervals too wide to tune against (a 14-positive and a 6-positive sample), plus **uncovered vendors at 6% of all sampled images** (千问/百度/星绘/抖音-class marks that no registered detector can ever fire on). Adding those vendors is now a larger win than any further tuning of the covered four, and `docs/watermarking-landscape.md` carries their researched specs.
+### Gemini sparkle
 
-**Per-mark geometry scaling (`scale_basis` / `scale_base`, 2026-07-18) -- the largest single recall defect found so far.** Every tuned fraction in `TextMarkConfig` was calibrated on PORTRAIT captures, where the width and the short side coincide, so the scaling basis was never exercised until landscape inputs were measured. Corpus-measured on 2572 unique TC260 carriers, BEFORE the fix:
+[`gemini_engine.py`](../src/remove_ai_watermarks/gemini_engine.py) uses a
+multi-scale shape search and a false-positive gate. Its captured sparkle assets
+serve detection and mask geometry only. Pixel recovery is performed by the
+shared fill backend.
 
-| aspect ratio | detected | missed | miss rate |
-|---|---|---|---|
-| tall portrait <0.70 | 323 | 212 | 40% |
-| portrait 0.70-0.95 | 607 | 533 | 47% |
-| square ~1.0 | 190 | 272 | 59% |
-| landscape 1.15-1.6 | 0 | 143 | **100%** |
-| wide >1.6 | 0 | 292 | **100%** |
+`detect_sparkle_confidence` uses a process-wide shared engine because its loaded
+assets and template ladder are immutable.
 
-**Not one landscape image in the corpus ever produced a detection** -- 435 of them, zero. A width-scaled box is inflated by the aspect ratio on a wide image, so the glyph never lands inside it and the blob never gets scored: of the 1452 no-detection TC260 images the median doubao NCC was **0.057**, with 49% at ~zero. This is a LOCALIZATION failure, not a threshold one -- only 2.7% of those images sat in the band a threshold change could reach, which is why a day of threshold tuning could never have found it. Re-running the previously-undetected set with a short-side basis recovers **56% of landscape** (12% square, 4% portrait; 20% overall).
+Regression coverage:
 
-**The basis is PER MARK because the vendors genuinely differ.** The same switch took jimeng's labelled landscape positives from 13/13 to **0/13**: the Jimeng wordmark tracks the WIDTH while the Doubao strip tracks the short side, even though both are ByteDance and share a corner. So doubao is `short`, jimeng is `width`, and samsung stays `width` because there is no corpus evidence either way (1 addition corpus-wide) and an unmeasured change is not an improvement. China's GB 45438-2025 clause 5.2(e) mandates glyph height >= 5% of "the shortest side", which is why short-side is the natural prior -- but jimeng's measured behaviour overrides the prior. Regression: `tests/test_text_mark_engine.py::TestScaleBasis`.
+- [`test_gemini_engine.py`](../tests/test_gemini_engine.py)
 
-**How this was missed for so long:** precision was measured repeatedly and recall never was. The eval harness now reports a `missed` column for exactly this reason, and it is what caught the jimeng regression the short-side switch introduced.
+### Text mark engines
 
-**Competitive detection among same-corner marks (`rivals` / `_rival_margin_ok`, 2026-07-18).** Detection was purely ABSOLUTE -- every engine scored its own template against its own threshold, so nothing ever asked the discriminative question "does this blob match the NEIGHBOUR's mark better than mine?". Doubao "豆包AI生成" and Jimeng "★ 即梦AI" both sit bottom-right in near-white CJK and survive the top-hat binarization as very similar blobs, so no absolute gate can separate them. Measured on hand-labelled examples, scoring BOTH templates against the SAME glyph blob (n=40 jimeng / 75 doubao / 20 other-vendor labels / 89 clean):
+[`_text_mark_engine.py`](../src/remove_ai_watermarks/_text_mark_engine.py)
+provides common localization, detection front ends, template caching, rival
+comparison, and footprint construction.
 
-| feature | separability (0.5 = useless, 1.0 = perfect) |
-|---|---|
-| absolute `ncc_jimeng` | 0.96 |
-| `ncc_jimeng` MINUS `ncc_doubao` | **0.99** |
+Each vendor module supplies a `TextMarkConfig` and only the behavior that cannot
+be represented by the shared base:
 
-At a 0.10 margin: real Jimeng wordmarks pass **100%**, Doubao strips 8%, other vendors' AI labels (千问/百度/星绘/抖音) 55%, no-mark corners 12%. Corpus effect (`scripts/visible_eval.py`, 741 labelled images): **jimeng precision 38% -> 63%, genuine detections unchanged at 40, false fires 65 -> 23.** Because real marks pass at 100% this is a pure precision gain, unlike raising a threshold -- so the earlier 0.85 relaxation patch was REVERTED to 0.70 and the recall it had sacrificed came back. **The gate is deliberately asymmetric:** doubao declares no rival, because the symmetric gate cost it 7 genuine detections to prevent 5 false ones (1.4:1 against) while jimeng gained 25pp for free -- doubao's absolute detector is already 86% precise and has nothing to buy. A rival's config is looked up lazily by asset name (`_rival_config`) so its template is scored at ITS own geometry; scoring it at the host mark's geometry would compare a correctly-sized template against a mis-sized one and hand the margin a free win. Regression: `tests/test_text_mark_engine.py::TestRivalMargin`.
+- [`doubao_engine.py`](../src/remove_ai_watermarks/doubao_engine.py)
+- [`jimeng_engine.py`](../src/remove_ai_watermarks/jimeng_engine.py)
+- [`qwen_engine.py`](../src/remove_ai_watermarks/qwen_engine.py)
+- [`kling_engine.py`](../src/remove_ai_watermarks/kling_engine.py)
+- [`yuanbao_engine.py`](../src/remove_ai_watermarks/yuanbao_engine.py)
+- [`samsung_engine.py`](../src/remove_ai_watermarks/samsung_engine.py)
+- [`runninghub_engine.py`](../src/remove_ai_watermarks/runninghub_engine.py)
+- [`baidu_engine.py`](../src/remove_ai_watermarks/baidu_engine.py)
+- [`liblib_engine.py`](../src/remove_ai_watermarks/liblib_engine.py)
 
-**Evaluation harness (`scripts/visible_eval.py` + `scripts/visible_groundtruth.py`, 2026-07-18).** Run before AND after any detector change; `--save NAME` snapshots, `--vs NAME` diffs. Ground truth is 741 blind-labelled corpus images (779 cells across two rounds, two-sided control each). Three properties of the harness are load-bearing and were each added after the naive version produced a wrong number:
+The detector and removal mask must use compatible geometry. A detector that
+fires while producing an empty or misplaced mask is a removal failure even if
+the detection test passes.
 
-* **Adjudication scope.** A crop centred on one mark only lets the labeller rule on marks visible IN THAT CROP. Scoring jimeng against a pill-round image (top-left crop) books real bottom-right detections as false fires -- ~61% of pills carry a wordmark. Each image records which marks its crop could rule on; bottom-right marks co-adjudicate each other.
-* **Provenance must come from METADATA, never from the labels.** A relaxation arm only fires when provenance names the vendor, so label-derived provenance hands the detector the answer: it scored gemini at 99% instead of the true 41%.
-* **Recall is NOT reported.** The labelled set was sampled where detectors fired, so images every detector missed are absent by construction; a recall computed here would divide by a denominator that excludes exactly the failures recall exists to expose. The `missed` column catches a change LOSING marks it used to find, nothing more. True recall needs a random corpus sample labelled exhaustively -- not yet done.
+Yuanbao uses the polarity-independent `contrast` front end because its standard
+two-line mark can be light on dark scenes or dark on light scenes. Its detector
+and footprint both use the same best-match box. The separate one-line overlay
+variant is not covered.
 
-Baseline at the time of writing (sensitivity `auto`, provenance from metadata): gemini 41% (321 fires), doubao 86% (77), jimeng 63% (63), jimeng_pill 64% (83), samsung unmeasurable (1 addition corpus-wide).
+The capture-less Jimeng pill lives in
+[`pill_engine.py`](../src/remove_ai_watermarks/pill_engine.py). It uses a
+synthetic silhouette for detection and a fixed top-left footprint.
 
-**Per-mark provenance NCC relaxation + the corroboration gate (2026-07-18).** Two defects, both on the DEFAULT `auto` path (no flag, driven by TC260 metadata), found by blind hand-labelling the ADDITIONS (accepted with provenance, rejected without) over 4417 unique TC260 carriers. Two-sided control: labeller sensitivity 100% (doubao) / 96% (jimeng), specificity 100% / 100% — the controls are what make the low numbers trustworthy, and the "clean" stratum is structural (another vendor's C2PA image, where a ByteDance mark cannot exist) rather than detector-defined, so it is not circular.
+Each engine has a corresponding test module under [`tests/`](../tests/).
+Shared behavior is covered by:
 
-(1) **One shared `_PROVENANCE_NCC_FACTOR = 0.7` meant two different things per mark:**
+- [`test_text_mark_engine.py`](../tests/test_text_mark_engine.py)
+- [`test_text_mark_faint_mask.py`](../tests/test_text_mark_faint_mask.py)
+- [`test_text_mark_memory.py`](../tests/test_text_mark_memory.py)
 
-| mark | band | precision | 95% CI | n |
-|---|---|---|---|---|
-| doubao | whole arm | 76% | 61-87% | 42 |
-| | [0.280,0.340) | 58% | 36-77% | 19 |
-| | [0.340,0.400) | 91% | 73-98% | 23 |
-| jimeng | whole arm | 17% | 10-27% | 82 |
-| | [0.315,0.383) | 12% | 6-22% | 68 |
-| | [0.383,0.450) | 43% | 21-67% | 14 |
+### Fill backends and region erasing
 
-The factor is now a per-mark `TextMarkConfig.provenance_ncc_factor`. Doubao stays 0.70 — both bands return more true marks than false fills, so tightening would cost 11 genuine recoveries to prevent 8. Jimeng moves to 0.85 (gate 0.3825), dropping the 12% band: −8 genuine recoveries, −60 false fills (7.5:1), arm precision 17% → 43%. **Why jimeng fails is a detector problem, not a threshold one:** of its 68 false additions, 33 were DOUBAO marks and 17 were other vendors' AI labels (千问 / 百度 / 星绘 / 抖音) — relaxed, the silhouette keys on "some text in the bottom-right corner", not on "★ 即梦AI". Damage was scored separately because doubao and jimeng share a corner: 45 of the 68 fill a corner nothing else would touch, the other 23 are harmless (doubao fires strictly there and fills the same box anyway). A better silhouette, not a lower factor, is the real fix.
+[`region_eraser.py`](../src/remove_ai_watermarks/region_eraser.py) implements the
+same backends used by visible removal and the user-directed `erase` command:
 
-(2) **A weak detector must not corroborate a sibling (`_CANNOT_CORROBORATE`).** `resolve_trust` grants `confirmed` on a strict-detected sibling of the same `_PRODUCT_OF`, and `confirmed` bypasses the sibling's FP gate outright. The pill (~7% documented raw false-fire; 5.5% on 578 vendor negatives) maps to product "jimeng", so it could hand that bypass to the wordmark — a closed loop: pill false-fires on clean non-ByteDance content → jimeng relaxes 0.45 → 0.3825 and false-fires → `_keep_pill` sees "jimeng" in keys and takes the WORDMARK arm, removing the pill **unrestricted**, skipping the flatness guard written to stop exactly that smear. 3 of 578 negatives ran the full loop, one with `footprint_flat=0`. The fix costs nothing: negatives 3 → 0, TC260 carriers unchanged (jimeng 398 → 398, pill 117 → 117). `_keep_pill` already encoded this distrust for the pill's ACTION; the gap was that its TESTIMONY was ungated.
+- `cv2`
+- `migan`
+- `lama`
 
-**Pill arms, re-measured on the same corpus** (149 blind-labelled TC260-arm fires, 35 wordmark, 33 unconfirmed): wordmark **94%** (CI 81-98%, confirming the original claim), TC260-metadata-only **21%** raw (CI 16-29%, consistent with the original ~27%) — **29%** (CI 20-40%) among the flat footprints the guard PASSES vs 14% among those it blocks. The guard works directionally but weakly: the shipped arm still runs at ~2.4 false fills per genuine one. Whether an arm that inaccurate belongs on the default path is a product call, not a tuning one.
+`watermark_registry.resolve_backend` selects LaMa first, then MI-GAN, then
+OpenCV for `auto`. A memory-constrained caller should explicitly select MI-GAN
+or OpenCV instead of relying on `auto`.
 
-**Samsung's relaxation is UNMEASURED and not measurable here:** the corpus holds 14 `samsung_genai` carriers and 3 visible Samsung detections total. Any precision estimate would carry a Wilson interval spanning most of [0,1]. Likely structural — detection is calibrated to the Italian locale string only.
+MI-GAN and LaMa crop around the mask before model inference and paste back only
+masked pixels. Their model sessions are loaded lazily. MI-GAN uses the inverse
+mask polarity expected by its ONNX model.
 
+Regression coverage:
 
-**Provenance prior:** when local metadata already confirms the vendor, the mark's detection trust gate is relaxed (a confirmed vendor means the mark is present with high prior, so a mark the conservative detector would demote as a content false positive is trusted). `detect_marks` / `remove_auto_marks` take a `provenance` frozenset and `KnownMark.remove` a `provenance` flag. Mapping: a Google/Gemini C2PA issuer relaxes gemini (skips its false-positive gate and lowers the trust threshold from 0.5 to 0.35); a China-AIGC (TC260) label relaxes doubao/jimeng; `samsung_genai` relaxes samsung. Corpus finding: on Google-C2PA images, Gemini sparkle recall rose from ~46% (plain detector) to ~90% with the provenance prior (recovering marks the vendor moved or re-rendered). That gain is why the bypass exists, and it is conditional on the metadata actually naming the vendor — a caller merely ASSUMING the image is AI does not get it unconditionally (see the assumed-trust confidence floor above). The localizer is cheap CPU (cv2/numpy), so a memory-tight caller runs it anywhere; the heavy MI-GAN/LaMa fill is opt-in and chosen by the caller.
+- [`test_region_eraser.py`](../tests/test_region_eraser.py)
+- [`test_inpaint_fallback.py`](../tests/test_inpaint_fallback.py)
 
-**Cross-engine confidences aren't directly comparable**, so the gemini adapter applies the corpus-validated 0.5 sparkle threshold (`_GEMINI_AUTO_MIN_CONF`) for its `detected` flag (lowered to 0.35 under the Google/Gemini provenance prior) — otherwise the gemini engine's loose internal threshold weakly fires (~0.36) on the Doubao text and hijacks `auto`. The shape-keyed Doubao/Jimeng/Samsung NCC detectors don't cross-fire (jimeng scores ~0.22 on the Doubao strip, well under its 0.45 threshold; Samsung is bottom-left so it shares no corner with the others, and scored 0.0 on Doubao/Jimeng captures and they 0.0 on a real Samsung photo), so `auto` picks the right one. `cli.cmd_visible` is registry-driven: `--mark auto` → `remove_auto_marks` (removes every detected mark), `--mark <key>` → that mark; `--mark` choices come from `mark_keys()`.
+## Invisible watermark regeneration
 
-**`cli._remove_visible_auto` is the shared visible-removal helper used by `cmd_all`/`cmd_batch` too** (they no longer hardcode `GeminiEngine`), so `all`/`batch` remove Doubao/Jimeng/Qwen/Samsung text marks, not just the Gemini sparkle (regression-guarded by `test_all_visible_step_uses_registry`). The three text-mark adapters were consolidated 2026-06-09: a single `_text_mark(key, label, location)` builds the registry row from one parameterized `_text_mark_detect`/`_text_mark_remove` pair (the remove adapter localizes the glyph footprint and hands it to the shared `fill` only when detected/forced, else skipped); the gemini adapters stay bespoke. Add a new visible mark = one `_text_mark(...)` row + its `TextMarkConfig` (with a captured alpha map for the detection silhouette); do not re-add per-mark `if` branches or copy-paste adapters.
+### Profiles and strength
 
-**Alpha-on-save policy (issue #30):** `image_io.write_bgr_with_alpha` (it lives in `image_io`, not `cli` — moved so the CLI and the library `api` share ONE implementation) rejoins the input's alpha plane **unchanged** — it must NOT zero alpha in the watermark bbox. The fill reconstructs real pixels there, so zeroing alpha punched a transparent hole that renders as a solid **white box** on any non-transparent viewer (Gemini app exports are opaque RGBA, so every user hit it; regression-guarded by `test_visible_keeps_alpha_opaque_in_watermark_region`). The registry `remove()` still returns its region, but the CLI no longer uses it to clear alpha. **It returns `imwrite`'s success flag and callers must check it** (2026-07-20): `imwrite` is contractually non-raising, so that bool is the only signal the file was not created. The wrapper previously returned `None` and swallowed it, so every CLI write site ran `output.stat()` to report the size and a read-only destination died with a bare `FileNotFoundError` traceback pointing at the stat instead of the write. The CLI now writes through the shared `cli._write_output_or_exit`. Regression: `tests/test_cli_robustness.py::TestFailedWriteIsReported`.
+[`noai/watermark_profiles.py`](../src/remove_ai_watermarks/noai/watermark_profiles.py)
+is the source of truth for:
 
-## `gemini_engine.py`
+- profile aliases;
+- default model identifiers;
+- default steps and seeds;
+- vendor-adaptive strength resolution;
+- the minimum viable step calculation.
 
-`gemini_engine.py` — visible Gemini-sparkle remover/detector (cv2/numpy, no GPU). `detect_sparkle_confidence(path)` is the file-level entry point used by `identify.py`. The public entry points normalize a grayscale (2D) or RGBA (4-channel) input to BGR up front so a non-BGR image does not crash the cv2 pipeline.
+The current profiles are `controlnet`, `sdxl`, `qwen`, and `qwen-zimage`.
+`default` is a legacy alias for `sdxl`. There is no content-dependent automatic
+router.
 
-**Detection localization (issue #36):** `detect_watermark`'s global multi-scale NCC search applies a size weight (`(scale/96)**0.5`) that suppresses tiny-patch false positives but can let a larger, mediocre match (e.g. a bright collar in a portrait) outrank a small, near-perfect sparkle in the corner — so a faint sparkle on a busy background scored below threshold and read as clean (the regression osachub reported from widening the search window 256px->512px between v0.7.2 and v0.8.8). `_corner_promote` adds a bottom-right-corner raw-NCC pass on top of the global search: a match with raw NCC >= `_CORNER_PROMOTE_NCC` 0.85 that beats the global pick overrides it (it only ever replaces a lower-fidelity pick, so it cannot weaken an existing detection), rescuing the buried sparkle without reverting the wider window. The corner side is **relative-clamped** (`_CORNER_PROMOTE_FRAC` 0.20 of the short side, clamped to `[_CORNER_PROMOTE_MIN` 96, `_CORNER_PROMOTE_MAX` 384`]`): a fixed 256px is a true corner on a large image but covers ~70% of a small portrait, where a real photo raw-matches the star at ~0.81 (relative tightening drops that worst case to ~0.69, while the upper clamp stops the corner ballooning on huge images where a real photo reached ~0.83 at 512px). The 0.85 gate sits midway between the worst real-photo corner match (~0.78 across native + downscaled negatives) and a genuine faint sparkle (~0.93), so promotion adds true detections with zero corpus false positives (Gemini's sparkle sits ~60-160px from the corner at fixed margins, covered by the [96, 384] band at every measured size). Regression-guarded by `test_gemini_engine.py::TestCornerPromotion`.
+[`invisible_engine.py`](../src/remove_ai_watermarks/invisible_engine.py) handles
+image sizing, optional pre-upscaling, postprocessing, and the public engine
+interface. It delegates model execution to
+[`noai/watermark_remover.py`](../src/remove_ai_watermarks/noai/watermark_remover.py).
 
-**Top-K fusion selection (osachub follow-up 2026-06-12):** `_corner_promote`'s 0.85 raw-NCC gate still missed a class the 256->512 widening exposed — a genuine MID-scale sparkle whose raw NCC sits *below* 0.85 but is buried by a LARGER, low-fidelity decoy that wins the size weight. The reporter's image (a scale-48 sparkle on light bedding) measured spatial 0.775 / grad 0.960 / fusion 0.676 at the true sparkle, but the size-weighted argmax instead locked onto a decoy at spatial 0.628 / grad 0.036 (fusion 0.325) — so `identify` read `unknown` on v0.8-0.11 where v0.7.2 (256px window) had caught it at 0.676. Fix: `detect_watermark` now keeps the **top-`_SELECT_TOPK` (3)** size-weighted candidates (NMS-deduped by location) plus the corner-promote candidate, scores EACH by the full fusion (spatial+gradient+variance) via the extracted `_grad_var_scores` helper, and selects the highest — the gradient term (the discriminator a contrast-invariant spatial NCC lacks) lifts the true sparkle over the decoy. Critically, selection ranks by the SIZE-WEIGHTED score, NOT raw NCC: a raw-NCC argmax (tried first) re-admitted the exact tiny-patch (scale 16-18) false positives the size weight exists to suppress — it flagged 14/65 doubao + 4/11 jimeng visible-corpus images (non-Gemini content) as Gemini sparkles. Top-K keeps tiny-patch suppression intact: a coincidental 16px match never ranks in the size-weighted top-K, so widening selection added **zero** flips on the doubao/jimeng corpora and left the 495-image Gemini set unchanged (479 detected, both before and after) while recovering the reporter's image. Regression-guarded by `test_gemini_engine.py::TestCornerPromotion::test_low_gradient_decoy_loses_to_high_gradient_corner_sparkle` (mirrors the real spatial/grad signature via a monkeypatched scan) and `test_size_weighted_search_alone_traps_on_the_decoy`.
+The Python engine and CLI do not have identical defaults for every optional
+postprocessing argument. Integrations that require reproducibility should pass
+the relevant values explicitly.
 
-**Square-image residual misses are NOT fixable by lowering the detector threshold (measured + REJECTED 2026-06-11):** osachub (#36 follow-up) reported the corner-promote still misses Gemini sparkles on Google **square (1:1)** outputs. Reproduced on the spaces corpus: of 330 square Google-C2PA images, 140 score below the identify 0.5 threshold, and visual review confirmed a real class -- faint white sparkles on dark/textured/colored backgrounds (raw NCC 0.46-0.73, below the 0.85 promote gate) landing at fusion conf 0.41-0.47. A margin-gated promote (promote when raw NCC >= 0.50 AND `_core_ring_margin` >= 40) rescued 32/33 confirmed misses at an apparent 0 FP, but that 0 was a **measurement artifact** -- the negative set was the margin<40 misses, which a margin>=40 gate excludes by construction. On an honest 518-image non-Google pool the same gate fired on **~174 (≈33%)**, visually content (screenshots, Chinese "AI生成" Doubao/Jimeng text marks, logos, bright textures), not sparkles. Adding an achromatic-core constraint (`chroma <= 15`) did not separate them either (kept 15/33 POS, 41 NEG still firing). Root cause is the documented contrast-invariant-NCC wall: a faint sparkle on a busy background is indistinguishable from a bright/ornate content corner at the (shape-NCC, brightness-margin, core-chroma) feature level.
+Regression coverage:
 
-**Conclusion: keep the 0.85 corner gate; do NOT add a margin/chroma-gated lower promote.**
+- [`test_watermark_profiles.py`](../tests/test_watermark_profiles.py)
+- [`test_invisible_engine.py`](../tests/test_invisible_engine.py)
+- [`test_img2img_runner.py`](../tests/test_img2img_runner.py)
+- [`test_platform.py`](../tests/test_platform.py)
 
-The cost (mislabel ~8-33% of non-Gemini content as Gemini) outweighs the benefit -- the visible sparkle is a medium-confidence stripped-metadata fallback, and intact Gemini is caught by C2PA in `identify` regardless. Remaining square misses are an accepted known limitation; a real fix would need a sparkle-specific discriminator (template match on a background-subtracted image, or a hard fixed-margin position prior), which is open research, not a threshold tweak.
+### CPU offload
 
-**Removal is localize -> fill** (`footprint_mask` → `watermark_registry.fill`): `footprint_mask` returns the sparkle footprint = the captured alpha (computed `alpha = max(R,G,B)/255` from the bundled sparkle-on-black captures `assets/gemini_bg_{96,48}.png`, capture max ~130 for the ~51%-opaque overlay) thresholded LOW so the faint halo is included, then dilated by a sparkle-relative margin. That binary mask is inpainted by the shared fill (cv2 / MI-GAN / big-LaMa). The captured alpha maps are used only to detect and to shape the mask, not for pixel recovery. This replaced the old reverse-alpha removal path; because the fill only reconstructs the masked footprint from its surroundings (rather than dividing by `1-a`), the whole reverse-alpha removal tail — the over-subtraction guard (`_reverse_alpha_oversubtracts`, the dark-background black-pit fix), the under-subtraction alpha-gain estimate (`_estimate_alpha_gain`), and the self-verify repair — is GONE, along with the near-white `1/(1-a)` ill-conditioning and the "color changed, not removed" failure mode those guards patched around. A slightly-off localization now just fills a small region near-losslessly instead of leaving a color-shifted smear.
+CPU offload is enabled only when requested on CUDA. The standard Diffusers
+profiles call `enable_model_cpu_offload`. The `qwen-zimage` profile uses the
+same flag to force its face stack out of automatic device residency.
 
-**False-positive gate (added 2026-06-03):** `detect_watermark`'s shape-only NCC (`spatial*0.5 + gradient*0.3 + var*0.2`) fires on ornate/flat content (text strips, banners, hatching) that coincidentally matches the diamond shape — a real Gemini sparkle is a bright WHITE overlay, so its core sits above the local background, but the NCC is contrast-invariant and cannot see that. The fusion now **demotes** (caps confidence to 0.30) any low-confidence (`< _SPARKLE_FP_CONF` 0.65) match that shows NEITHER real-sparkle signature: a bright core (`_core_ring_margin >= _SPARKLE_FP_MARGIN` 5) OR a crisp star silhouette (`gradient_score >= _SPARKLE_FP_GRAD` 0.55). I.e. demote when `low_margin OR low_grad`. Real sparkles escape via high confidence (white-bg sparkles score ≥0.79 despite a low margin — the NCC shape match is strong), high margin (dark/mid backgrounds, incl. the #36 faint-corner case, lift well clear), OR high gradient (a real sparkle is grad ~0.97–1.0). **The gradient condition (added 2026-06-26) closes the bright-background FP class** the margin check alone missed: a snow+sky photo and a white-background product render both scored ~0.51 at `identify`, because a bright background gives the match a HIGH core-ring margin (it genuinely IS brighter than its surroundings), so the brightness gate read it as a real overlay — but a smooth luminance blob that shape-NCC-matches the rough diamond has low gradient fidelity (the two FPs measured grad 0.105 and 0.463 vs ≥0.8 for real sparkles), so the gradient floor demotes them. The OR is **strictly a superset** of the old margin-only demotion (it only ADDS demotions on bright backgrounds, where a real sparkle keeps grad ~0.97), so it cannot regress a dark/mid sparkle (kept by margin) or a white-bg one (kept by confidence ≥ 0.65). The gate is **monotonic** (only ever removes detections, never adds), so it cannot regress the verified-negative corpus (already 0 FPs); the 2026-06-26 corpus re-sweep flipped only OpenAI/ChatGPT content (no Gemini sparkle exists there) and already-`cleaned/` outputs, all sub-0.5 (below the `identify` threshold), so no provenance verdict changed. The original gate demoted 16/495 flagged sparkles on the validation corpus (13 carried no AI metadata = content FPs; the 3 AI-meta were visually FPs / a near-invisible white-on-white sparkle whose AI verdict is held by metadata anyway). `_core_ring_margin` uses the `_core_and_bg` helper (core 75th-pct brightness vs background-ring median). This gate is detection-side and unchanged by the localize -> fill refactor; the provenance prior skips it when a Google/Gemini C2PA issuer confirms the vendor. Regression-guarded by `test_gemini_engine.py::TestSparkleFalsePositiveGate` (incl. `test_bright_background_low_gradient_match_demoted`).
+Regression coverage:
 
-**The reverse-alpha removal tail is retired.** The self-verify repair (`_verify_and_repair`), the offset+scale alignment search, and the near-white `1/(1-a)` ill-conditioning survivors were all artifacts of solving the sparkle by inverting the alpha map. Under localize -> fill the footprint is reconstructed from its surroundings by the shared fill, so those failure classes and their guards no longer exist. The lesson from that era still holds and generalizes: a re-detect-confidence audit metric is gameable by reshaping the residual, so judge a visible removal by physical inspection of the footprint, not the detector alone.
+- [`test_cpu_offload.py`](../tests/test_cpu_offload.py)
 
-**The bg assets are rebuilt from OUR OWN controlled captures** (`data/gemini_capture/captures/`, committed) by `scripts/visible_alpha_solve.py gemini`, which locates the 96px sparkle on the black capture and crops it to the two logo sizes; our capture matched the previously third-party-sourced `gemini_bg_96.png` to **NCC 0.9998**, validating the asset and making it reproducible. Gemini's multi-size fixed-slot model is genuinely different from the Doubao/Jimeng text-strip engines (so it stays a separate engine, not part of the shared-base refactor).
+### Qwen plus Z-Image
 
-## `_text_mark_engine.py`
+[`noai/qwen_zimage_pipeline.py`](../src/remove_ai_watermarks/noai/qwen_zimage_pipeline.py)
+implements the fixed CUDA-only two-stage profile:
 
-`_text_mark_engine.py` — **shared base for the registered text-mark engines, extracted 2026-06-09** (they were ~90% byte-identical clones). `TextMarkEngine(config: TextMarkConfig)` owns the `locate → extract_mask → detect` detection pipeline plus the removal that localizes the glyph blob to a footprint mask and hands it to the shared `watermark_registry.fill` (+ the asset-keyed `load_alpha_template`/`glyph_silhouette`/`template_match_score` caches). Detection still matches the glyph silhouette (NCC against the captured template); the removal MASK is TEMPLATE-FREE — it is the bounding box of the top-hat glyph blob from `extract_mask`, filled solid + dilated, so a re-rendered or differently-placed mark is still masked. This dropped the fixed alpha-template placement; the captured alpha maps are now used only for the detection silhouette, not for removal. Each engine module is a thin subclass supplying only its `TextMarkConfig` (the tuned constants, the bundled asset, and the bounded structural deltas — `corner` br/bl, `margin_floor` 4/2, `morph_open_size` 5/3, `min_gw` 8/16, and since 2026-07-21 `ladder` — the scale rungs `_tophat_best` sweeps, per-mark because 千问's two size modes do not fit the shared 3-rung comb (default `(0.8, 1.0, 1.25)`, unchanged for every other mark; densifying the SHARED ladder was measured and rejected -- see the verification plan's B2). plus the test-facing module shims (`_alpha_template`/`_glyph_silhouette`/`_template_match_score` + the constants). Gemini stays a SEPARATE engine (its multi-size fixed-slot sparkle model is genuinely different). Add a new text mark = a new `TextMarkConfig` + a thin subclass + one registry `_text_mark(...)` row. The engine bullets below describe each mark's calibration history; the LOGIC lives here. **Small-image detection guard (`_MIN_DETECT_SHORT_SIDE` 200, added 2026-06-26):** `detect` returns not-detected when the image short side is below 200px. Below that the glyph template degrades to the `min_gw` floor (~8px) and `TM_CCOEFF_NORMED` on a few pixels is noise, so an unrelated small geometric shape can spuriously correlate with the CJK silhouette — a 48×48 app-icon chevron scored Doubao 0.41 / Jimeng 0.47 (both above their thresholds), a pure small-size artifact (the same icon upscaled collapses to ~0.06–0.10 NCC at ≥256px). A real AI-generation label is stamped on a full-resolution render (the captured samples are 1086–2048px wide, the smallest positive test image is 1086px), so the floor sits far below any genuine mark while killing the icon/thumbnail band (≤96px); `identify` falls back to "unknown" (the safe default) and removal, gated on detection, is suppressed too. Regression-guarded by `test_{doubao,jimeng,samsung}_engine.py::TestDetect::test_small_image_guarded_from_false_positive`.
+1. Qwen Image with Canny conditioning regenerates the frame.
+2. YuNet locates faces, SAM builds masks, and Z-Image regenerates the selected
+   face regions.
 
-**Removal is localize -> fill.** The engine localizes the glyph blob (`extract_mask` over the located box) into a solid, dilated footprint mask and hands it to the shared `watermark_registry.fill` (cv2 / MI-GAN / big-LaMa). The template-free mask (bounding box of the glyph blob, not the fixed alpha template) means a re-rendered or moved mark is still covered, and the fill reconstructs the box from its surroundings. On corpus images doubao and jimeng localize + remove at ~100% with clean footprints (the filled region blends into its surroundings within a few LAB levels, no color shift, no dark pit); clean images with no vendor signature had 0% false removal.
+The profile rejects a custom model identifier. Its global and face model stack
+is fixed by the implementation. When tiling is enabled, only the global stage
+is tiled; the face stage runs once after the tiles are blended.
 
-**The reverse-alpha removal machinery is retired.** The old per-glyph reverse-alpha blend (`_apply_reverse_alpha`), the fixed/aligned alpha-map helpers, the over-subtraction guard (`_reverse_alpha_oversubtracts` → `_inpaint_footprint`, the dark-pit fix on dark/mid-tone backgrounds), and the always-align placement search are all gone — the fill reconstructs the footprint from its surroundings rather than inverting the captured alpha, so the dark-pit and color-shift failure modes those guards patched around no longer arise. `extract_mask` still returns a box-sized (`(loc.h, loc.w)`) mask rather than a full frame, which keeps the memory-tight `identify` detect path cheap.
+Regression coverage:
 
-**Polarity-independent contrast front-end (`detect_frontend="contrast"`, 2026-07-25).** Some vendors choose light or dark text from the scene under the mark, so a white top-hat cannot represent both. The contrast response subtracts a local Gaussian luma estimate, takes the absolute residual, suppresses saturated pixels, and max-normalizes before silhouette NCC. One `_contrast_best` method supplies both the score and the winning match box; `footprint_mask` uses that same box after a successful detection, preserving the detector-to-mask parity contract. This mode was added for Tencent Yuanbao and its 0.38 gate is specific to that response. It must not be copied to another front-end or mark without recalibration.
+- [`test_qwen_zimage_pipeline.py`](../tests/test_qwen_zimage_pipeline.py)
+- [`test_cpu_offload.py`](../tests/test_cpu_offload.py)
 
-## `doubao_engine.py`
+### Tiling
 
-`doubao_engine.py` — **a thin `_text_mark_engine.TextMarkEngine` subclass (config only) since 2026-06-09.** visible Doubao "豆包AI生成" detector + localizer (cv2/numpy, no GPU). `DoubaoEngine.locate` anchors a bottom-right box by **geometry** (mark scales with image WIDTH), `extract_mask` pulls the light, low-chroma glyphs (the detection candidate) using a per-pixel channel-spread proxy `sat = roi.max(axis=2) - roi.min(axis=2)` (no HSV conversion). `detect` is **shape-consistent**: it matches the bundled glyph silhouette (`assets/doubao_alpha.png`) against the candidate via zero-mean normalized correlation (`_template_match_score`, cv2 `TM_CCOEFF_NORMED`), gated at `DETECT_NCC_THRESHOLD` 0.4 over a small `DETECT_MIN_COVERAGE` floor. Keying on glyph SHAPE (not coverage heuristics) fixed #23 (corpus FP 7/1243).
+[`noai/tiling.py`](../src/remove_ai_watermarks/noai/tiling.py) contains pure
+tile planning, feather weights, tile orchestration, and region compositing.
 
-**Removal is localize -> fill:** the glyph blob is localized to a solid, dilated footprint mask (`extract_mask` over the located box) and the shared `watermark_registry.fill` inpaints it. On corpus images this removes at ~100% with clean footprints (the filled region blends into its surroundings within a few LAB levels, no color shift, no dark pit).
+Tiling engages only when requested and the long side exceeds the tile size.
+It avoids an explicit full-image downscale but does not make diffusion
+pixel-preserving. Each tile is still regenerated.
 
-**The detection template (`assets/doubao_alpha.png`) is rebuilt by `scripts/visible_alpha_solve.py`** (the careful gray-self solve: cubic background fit, mean over channels, full halo, unblurred), same recipe as Jimeng — the captures are committed in `data/doubao_capture/captures/`. It is used only as the detection silhouette, not for pixel recovery.
+`feather_region_composite` changes only the requested box and leaves pixels
+outside it unchanged.
 
-**The locate box (`WM_*`) is generous (0.22 wide, margins 0.004) and reaches close to the corner** so a re-rasterized, corner-ward-shifted mark still falls inside the localized box; regression-guarded by `test_recovers_shifted_mark_on_texture` (composes the mark shifted on a known texture). **`extract_mask` guards a degenerate ROI (`bh < 16 or bw < 16` -> empty mask, skips cv2)** — an extremely wide/short image (e.g. 2048x1, `test_wide_short_does_not_raise`) once fed cv2's GaussianBlur a ~1-px-tall ROI and **faulted natively on Windows py3.12**; real images always clear the guard (the `WM_*` box floors are `max(16, …)` height / `max(40, …)` width), so it only short-circuits slivers. The registry gates removal on `detect`. The shipped third-party `_refs/zhengsuanfa_doubao_alpha_120x20.png` is NOT a usable template (verified 2026-05-29). Arbitrary-region inpainting is `region_eraser`/`erase`. **Lesson from the reverse-alpha era (still holds): a detector-only removal test is insufficient; assert visual residual (the textured-shift test).**
+Regression coverage:
 
-## `jimeng_engine.py`
+- [`test_tiling.py`](../tests/test_tiling.py)
 
-`jimeng_engine.py` — **a thin `TextMarkEngine` subclass (config only) since 2026-06-09.** visible Jimeng / Dreamina "★ 即梦AI" detector + localizer (cv2/numpy, no GPU), built 2026-05-30 from issue #13's solid captures (@powersee). Shares the base with `doubao_engine`: `locate` anchors a bottom-right box by **geometry** (scales with WIDTH), `extract_mask` pulls the light low-chroma glyphs (white top-hat + grayish + min-luma), `detect` matches the bundled "即梦AI" glyph silhouette (`assets/jimeng_alpha.png`) via `TM_CCOEFF_NORMED` over a coverage floor. Threshold `DETECT_NCC_THRESHOLD` **0.45** cleanly separates real Jimeng marks (>=0.81) from the Doubao strip (0.21) and other AI output (0.0), so the two ByteDance marks don't cross-fire in `--mark auto`.
+### Upscaling and postprocessing
 
-**The detection template (`assets/jimeng_alpha.png`) is rebuilt by `scripts/visible_alpha_solve.py` from the GRAY capture** (`data/jimeng_capture/captures/`, the solid captures committed): `a = (I - B)/(255 - B)`, B a per-capture **cubic** background fit over the non-glyph pixels, **averaged over channels, full halo extent (down to a~0.02), unblurred**. Gray (bg ~132) is the deliberate choice over black: it is the best proxy for real content (the mark sits on bright photo areas, not on black). The captured template is used only as the detection silhouette, not for pixel recovery. Solver geometry at `_ALPHA_NATIVE_WIDTH` 2048: `_ALPHA_WIDTH_FRAC` 0.202, `_ALPHA_HEIGHT_FRAC` 0.058, margins ~0.029.
+[`upscaler.py`](../src/remove_ai_watermarks/upscaler.py) is the optional
+Real-ESRGAN path used only when enlarging a small image to the minimum
+resolution floor. Failure or an absent extra falls back to Lanczos.
 
-**Removal is localize -> fill:** the glyph blob is localized to a solid, dilated footprint mask and the shared `watermark_registry.fill` inpaints it; the `WM_*` locate box is generous so a re-rasterized, corner-ward-shifted mark stays inside the localized box (the same widen that fixed Doubao). On corpus images this removes at ~100% with clean footprints (blends within a few LAB levels, no color shift). The registry gates removal on `detect`.
+[`humanizer.py`](../src/remove_ai_watermarks/humanizer.py) contains explicit
+grain, unsharp masking, and adaptive polish helpers.
 
-**No committed real sample** (only the solid calibration captures are committed) — `tests/test_jimeng_engine.py` synthesizes a mark from the bundled template, and `test_recovers_shifted_mark_on_texture` guards the localize-on-shift path that the Doubao defect exposed. Jimeng images are independently caught by the China TC260 AIGC label in `metadata`/`identify`, so this engine is the visible-mark *removal* path, not a new `identify` signal.
+Regression coverage:
 
-## `samsung_engine.py`
+- [`test_upscaler.py`](../tests/test_upscaler.py)
+- [`test_humanizer.py`](../tests/test_humanizer.py)
 
-`samsung_engine.py` — **a thin `TextMarkEngine` subclass (config only) since 2026-06-09.** visible Samsung Galaxy AI "✦ Contenuti generati dall'AI" detector + localizer (cv2/numpy, no GPU), built 2026-06-05 from issue #37's flat captures (@f-liva). Shares the base but anchored **bottom-LEFT** (Doubao/Jimeng are bottom-right): `locate` anchors a bottom-left box by **geometry** (scales with WIDTH), `extract_mask` pulls the light low-chroma glyphs (white top-hat + grayish + min-luma — `LOGO_MIN_LUMA` is lowered to **110** because the mark is faint, peak alpha ~0.38, so on a mid/dark background its glyph luma is lower than Jimeng's), `detect` matches the bundled glyph silhouette (`assets/samsung_alpha.png`) via `TM_CCOEFF_NORMED` over a coverage floor. Threshold `DETECT_NCC_THRESHOLD` **0.40** (real marks ~0.79 on a real photo, ~0.57/0.71 on the black/gray captures; 0.0 on Doubao/Jimeng captures, and Doubao/Jimeng score 0.0 on a real Samsung photo — no cross-fire, also because the corner differs).
+## Image input and output
 
-**The detection template (`assets/samsung_alpha.png`) is solved by `scripts/visible_alpha_solve.py samsung` from the GRAY capture** (`data/samsung_capture/captures/`, the flat black/gray/white captures committed; the solver gained a `corner="bl"` mode + left-margin logging for this), same careful recipe as Jimeng (cubic background, mean-channel, full halo, unblurred). Geometry emitted at `_ALPHA_NATIVE_WIDTH` **1086** (the flat-edit capture width): `_ALPHA_WIDTH_FRAC` 0.3195, `_ALPHA_HEIGHT_FRAC` 0.0378, `_ALPHA_MARGIN_LEFT_FRAC` 0.0110, `_ALPHA_MARGIN_BOTTOM_FRAC` 0.0064. Used only as the detection silhouette, not for pixel recovery.
+[`image_io.py`](../src/remove_ai_watermarks/image_io.py) is the shared image
+codec boundary.
 
-**Removal is localize -> fill:** the glyph blob is localized to a solid, dilated footprint mask and the shared `watermark_registry.fill` inpaints it. Verified on a real 2958-wide @f-liva photo: re-detect 0.79→0.00, no readable text or outline on the recovered wooden table — checked **visually**, not just by the detector. The registry gates removal on `detect`.
+Contracts:
 
-**Detection is locale-specific** (the string differs per language); this build detects only the Italian "Contenuti generati dall'AI" variant, so non-Italian Samsung locales are not detected — and, because detection gates removal, not removed — even though the fill mask itself is locale-independent. Other locales need their own detection silhouette — the locale string font-rendered and calibrated on real positives (the pill's `scripts/render_pill_silhouette.py` pattern), NOT an app capture (the solid/gray/white capture workflow retired with reverse-alpha). This is a pre-existing limit, unchanged by the localize -> fill refactor.
+- All package OpenCV file reads and writes use `image_io.imread` and
+  `image_io.imwrite`.
+- `to_bgr` normalizes grayscale and alpha-bearing arrays.
+- `read_bgr_and_alpha` and `write_bgr_with_alpha` preserve the alpha plane.
+- `imwrite` returns a success flag; every caller must check it.
+- HEIC, HEIF, and AVIF fall back to Pillow plus `pillow-heif`.
+- A visible no-op can preserve the original file bytes.
 
-**No committed real sample** (only the flat calibration captures are committed) — `tests/test_samsung_engine.py` synthesizes a mark from the bundled template (bottom-left geometry), with `test_recovers_shifted_mark_on_texture` guarding the localize-on-shift path. Samsung Galaxy AI edits are independently caught by C2PA + the `genAIType` marker in `metadata`/`identify`, so this engine is the visible-mark *removal* path; it also feeds `identify` as the medium-confidence `visible_samsung` signal via the registry (the stripped-metadata fallback).
+Regression coverage:
 
-## `qwen_engine.py`
+- [`test_image_io.py`](../tests/test_image_io.py)
+- [`test_cli_robustness.py`](../tests/test_cli_robustness.py)
 
-`qwen_engine.py` — **a thin `TextMarkEngine` subclass (config only), registered 2026-07-21.** visible Qwen (Alibaba Tongyi Qianwen) "千问AI生成" detector + localizer (cv2/numpy, no GPU), bottom-right, the same GB 45438-2025 6-glyph house style as Doubao (2-glyph vendor prefix + mandated `AI生成` tail; the vendor's tri-lobe logo precedes the text and is deliberately NOT in the silhouette -- logos vary between releases, the CJK run is what discriminates). The detection silhouette `assets/qwen_alpha.png` is font-rendered synthetic (`scripts/render_vendor_silhouettes.py`), never cut from an upload. Feeds `identify` as the medium-confidence `visible_qwen` signal via the registry.
+## Adding or changing behavior
 
-**Why this registration took two attempts:** the 2026-07-18 attempt died at n=1 positives; the unlock was the TC260 label's `ContentProducer` field, whose USCC names the signing entity and partitions carriers into per-vendor cohorts from metadata alone (`scripts/vendor_cohort_harvest.py` -- 117 labelled Qwen frames, owing nothing to any pixel detector). Every constant was then MEASURED on that cohort against 286 hand-labelled clean frames (`scripts/vendor_mark_calibrate.py`), not inherited from Doubao:
+For a new visible mark:
 
-* **Basis `short`** (frac_short CV 0.189 vs width 0.273).
-* **Per-mark 2-rung `ladder=(0.78, 1.27)`** against `alpha_width_frac` 0.160: the mark sits in TWO size modes (~0.124 and ~0.203 of the short side, ratio 1.64, both clusters tight) -- wider than the shared 3-rung ladder's 1.5625 span, so the best single fraction covers only 74.5% and the small mode lands in the comb's collapse zone (real marks at the exact rung score ~0.94 vs ~0.67 on the shared ladder). A 4-rung variant scored strictly worse (its big-mode rung sits 4.6% off the mode). The shared default is untouched for every other mark.
-* **Fitted locate box** (`width_frac` 0.231, `height_frac` 0.074, margins ~0.021): the real mark sits ~0.025 of the short side off the right edge, and Doubao's box (0.004 margin) clipped the first glyph -- an exact-size template collapsed 0.73 -> 0.26 on a real frame.
-* **`alpha_height_frac` 0.0416 from the aspect fit** (p50 aspect 0.260 at the winning width) -- not the silhouette's own aspect (0.2219) and not Doubao's ratio.
-* **Gate 0.45**: clean p99 0.301 / max 0.316 (286 frames), and every cohort frame >= 0.45 carries a visible mark (83 of ~96 eyeballed visible marks fire = 86% recall of visible marks; the misses are white-on-near-white contrast losses). 0.45 was picked over 0.32 for margin against unseen clean content at zero measured recall cost.
-* **STRICT ONLY (`provenance_ncc_factor` 1.0):** the score band just below the gate is dominated by non-Qwen banners on same-cohort frames (a 夸克 anti-forgery strip at 0.274, a 造点 mark at 0.253), so a provenance-relaxed arm would be mostly false fills. There is no provenance mapping for qwen and no relaxed arm.
-* **No rival margin:** 0 cross-fires on 400 Doubao-marked / 298 Jimeng-marked / 286 clean frames at the gate, while a 0.10 margin would have cost ~10% of genuine Qwen detections. A confident Qwen detection suppresses the Jimeng pill exactly like Doubao's does (`_keep_pill` -- a Qwen frame is TC260 too but is not Jimeng-basic).
+1. create a synthetic detection silhouette;
+2. add or extend a vendor engine;
+3. add one registry entry;
+4. test detection, false positives, localization, and actual pixel change;
+5. update [supported signals](supported-signals.md).
 
-**Parity:** detect -> cv2 fill -> re-detect is clean on **83/83** real cohort marks, no empty masks (the `tophat` faint-mask fallback rides the same `_tophat_best`); the e2e suite drives one live cohort positive through the real CLI (`scripts/real_examples_e2e.py`, qwen bucket = symlinks under the gitignored `_visible_datasets/`). Regression: `tests/test_qwen_engine.py` (pins the ladder, the strict-only factor, the box anchor, and both size modes at a score floor that discriminates the shared-ladder and Doubao-margin mutations).
+For a new metadata signal:
 
-**The clean-arm contamination trap (load-bearing for any future calibration):** the 2026-07-18 `present: []` labels are in the vocabulary of the REGISTERED marks only, so 146 of the 432 "clean" frames sit in a TC260 cohort -- including Qwen-cohort frames visibly carrying 千问AI生成. They made up the clean arm's entire top tail (clean p99 0.37 -> 0.69). `vendor_mark_calibrate.load_sets` now excludes every frame in ANY TC260 cohort from the clean arm; a gate read off the unguarded arm is meaningless.
+1. add the scanner;
+2. add every supported removal placement;
+3. verify the output through `strip_and_verify`;
+4. add identification and removal tests;
+5. update [supported signals](supported-signals.md) and, when relevant,
+   [the watermarking landscape](watermarking-landscape.md).
 
-## `yuanbao_engine.py`
+For a diffusion change:
 
-`yuanbao_engine.py` — **thin `TextMarkEngine` subclass, registered 2026-07-25.** Tencent Yuanbao's standard visible mark is a compact italic two-line block, `元宝` over `AI生成`, at the bottom-right. It feeds `identify` as `visible_yuanbao`.
-
-* **The earlier measured negative was invalidated at the renderer.** A negative shear was applied without an x translation, clipping much of the lower `AI生成` line off the left edge while retaining a wide blank tail. The matcher then squeezed that malformed asset into the fitted geometry, so the old "no separation" result measured a renderer bug rather than the mark. `render_vendor_silhouettes.py` now translates before negative shear and tightly crops the result.
-* **Polarity-independent `contrast` front-end:** real Yuanbao frames switch between light-on-dark and dark-on-light stamps. Absolute local-luma residual preserves both forms, while the white top-hat misses the dark one. The synthetic silhouette uses Hiragino Sans GB W6, tight line spacing, a two-pixel dilation, and -0.60 shear.
-* **Measured gate and anchor:** gate **0.38**, STRICT ONLY, plus a bottom-right anchor requiring both margins at most 0.04 of the short side. Across 33 byte-unique cohort frames, 28 carry the standard two-line mark and 26 fire (92.9% recall of that variant). The guarded clean arm is 0/286 fires, maximum score 0.348.
-* **Removal parity:** detect -> cv2 fill -> re-detect is clean on 26/26 detected real marks. The detector's own match box is the footprint, with median full-frame area 0.70% and maximum 0.95%.
-* **Known variant limit:** one frame carries a separate single-line photographer overlay rather than the standard two-line stamp. It stays unregistered because one example cannot support a recall or false-positive calibration.
-
-Regression: `tests/test_yuanbao_engine.py` covers both polarities, the anchor, the footprint, registry wiring, and detector-clean removal. A confident Yuanbao detection suppresses the Jimeng pill because both can occur on TC260-labelled content but Yuanbao is not Jimeng-basic.
-
-## `runninghub_engine.py`
-
-`runninghub_engine.py` — **thin `TextMarkEngine` subclass, registered 2026-07-22.** RunningHub (hosted ComfyUI platform, USCC 91340100MAEB4N8H76, 73-frame cohort) "RunningHub AI生成" detector + localizer, **top-left** (the first `corner="tl"` mark), faint mid-gray latin+CJK text. Feeds `identify` as `visible_runninghub`.
-
-* **`gray` front-end (the third one, added for this mark):** the mark's faint gray is suppressed by the white top-hat to clean-arm levels (positives 0.16-0.23 vs clean p99 0.31), while raw-grayscale silhouette NCC separates (positives 0.38-0.54 vs clean p99 0.264 / max 0.304 on 283 guarded clean frames). It is contrast-DEPENDENT, unlike tophat -- one method `_gray_best` serves both detection and the mask, same one-method parity contract as `_tophat_best`.
-* **Tight ladder (0.95, 1.0, 1.05) exactly on the measured 0.32-of-width:** the NCC comb is razor-sharp in size (0.537 on-size, 0.223 at +5.6% off -- Qwen's comb behaviour, measured again here), so the shared 3 rungs (nearest rung 5.6% off) collapsed the match to 0.22 and the first calibration showed no separation at all.
-* **Anchor gate in `detect`:** the full-corpus sweep (`data/spaces/_sweep_new_marks.py`, 42009 files) fired on 37 outside-cohort frames at 0.34-0.38 (hair tops, shelves, window frames, CJK banners) -- no NCC threshold separates them from the 0.381 positives. Every positive sits at the measured corner (x 0.008-0.014, y 0.005-0.007 of the frame) and every false fire off it (x 0.013-0.150, y 0.009-0.045), so detection additionally requires the match box inside x<=0.025 / y<=0.015 of the frame. 0/37 false, 4/4 positives kept.
-* **Footprint is always the detector's match box** (never the binary blob): the blob under-segments the faint head glyphs and left "Runni" unremoved (caught visually on the first removal). Gate 0.34, STRICT ONLY. Regression: `tests/test_runninghub_engine.py`.
-
-## `baidu_engine.py`
-
-`baidu_engine.py` — **thin `TextMarkEngine` subclass with a custom `footprint_mask`, registered 2026-07-22.** Baidu (USCC 91110000802100433B, 16-frame cohort) "百度 AI生成" detector + localizer, bottom-right: a white bold 百度 text run + a separate white rounded tag with dark "AI生成". Feeds `identify` as `visible_baidu`.
-
-* **Detection keys on the 百度 text run ONLY.** A two-component (text+pill) template was measured and rejected: the solid white pill is a bright-blob magnet and both front-ends scored the clean arm at cohort levels (tophat clean p95 0.445 / gray clean p95 0.487 vs cohort ~0.5). The text-only silhouette separates (cohort 0.39-0.65 vs clean max 0.352). The white tag is still removed: the footprint extends to the corner (below).
-* **Two load-bearing rival margins** (`rivals=("doubao_alpha.png","qwen_alpha.png")`): 百度 vs 豆包 share their second glyph, and 百度 vs 千问 are near-identical after binarization -- at the 0.37 gate the template fired on 45.8% of 400 Doubao-marked frames and, on the 741-frame blind-labelled eval set, on 12 Qwen-marked frames at 0.38-0.43. Doubao's template beats it by ~0.56 on Doubao marks, Qwen's by 0.17-0.35 on Qwen marks; the 0.10 margin suppresses all crossfire at zero genuine-Baidu cost (cohort fire+m == fire).
-* **Gate history, each step measured:** 0.37 from the clean arm (max 0.352) -> 0.43 after the eval-set crossfires (the one 抖音 AI创作 fire at 0.425 named no registered rival) -> **0.48** after the full-corpus sweep put outside-cohort TRUE carriers at 0.50-0.66 vs the false arm max 0.47 (大众点评 UI, a math blackboard, an 80s banner). Cohort keeps 7/16 (all true); the sweep also found 6 metadata-STRIPPED true Baidu carriers the TC260 cohort cannot see -- the direct evidence that registration pays beyond the cohort.
-* **Custom `footprint_mask`:** the tag's flat white interior gives no top-hat response (a top-hat answers edges, not flats), so the base blob bbox ended at the text run and the fill left the tag as a ghost. The mask is the detector's match box extended RIGHT to the corner. STRICT ONLY. Regression: `tests/test_baidu_engine.py`.
-
-## `liblib_engine.py`
-
-`liblib_engine.py` — **thin `TextMarkEngine` subclass with a custom `footprint_mask`, registered 2026-07-22.** LibLibAI (哩布哩布AI, USCC 91110105MACJ6K1C8A, 15-frame cohort) triangle logo + "LibLibAI" wordmark detector + localizer, **bottom-CENTER** (the first `corner="bc"` mark; the locate box is horizontally centered). Feeds `identify` as `visible_liblib`.
-
-* **The discriminative lever is the silhouette FONT.** With the CJK house font (STHeiti) the cohort scored 0.31-0.47 against a full-corpus false arm (latin UI text bands, website screenshots) at 0.50 -- no separation at any gate. Measured across 7 candidate fonts, **Arial** lifts the cohort to 0.42-0.73 and DROPS the false arm to max 0.398: generic latin text matches the wrong font less, which is where the discrimination comes from. Gate 0.42 keeps all 8 marked cohort frames (0.43-0.59).
-* **Per-mark size floor (`_MIN_SHORT_SIDE=480`):** the one false fire with the final template was a 200x200 icon (0.444, on a 20px template). The shared `_MIN_DETECT_SHORT_SIDE` (200) is a crash guard, not a discrimination floor; the template needs ~48px to discriminate.
-* **Custom `footprint_mask`:** the base blob bbox was wrong in both directions -- it bled UP into background structure (ate a shirt's real print on the 768x1024 cohort frame) and never owned the triangle logo. The mask is the detector's match box extended LEFT by ~1.3 glyph heights (the logo is ~1.0x the glyph height, gap ~0.3x, measured on the cohort zoom). STRICT ONLY. Regression: `tests/test_liblib_engine.py`.
-
-## `region_eraser.py`
-
-`region_eraser.py` — universal region eraser (`erase` CLI) AND the shared fill backend behind `watermark_registry.fill` for the visible localize -> fill removal. `erase(image, boxes=|mask=, backend=)` accepts grayscale (2D) and RGBA (4-channel) inputs on **all** backends (each splits off any alpha plane and re-attaches it unchanged, and promotes grayscale to BGR): `boxes_to_mask` → one of three backends.
-- `cv2` (default, no deps): `cv2.inpaint`.
-- `migan` (extra `migan`, `andraniksargsyan/migan` ONNX, MIT, ~28 MB): `erase_migan`. Like `erase_lama`, it crops a padded region around the mask (`pad = max(256, 2*bbox)`), feeds only that crop to the ONNX model, and pastes only masked pixels back — but since MI-GAN accepts arbitrary dims (unlike LaMa's fixed 512² square) the crop is fed at NATIVE resolution (no resize). This **bounds the ONNX working set by the mark size, not the image**: feeding the whole frame made peak RAM scale with the upload (~0.6 GB at 4 MP up to ~2.4 GB at 25 MP, measured 2026-07); cropping holds it roughly constant (~0.6-0.9 GB), so a memory-tight host (a 1-2 GB web worker) can run MI-GAN on a 25 MP upload. The crop does not degrade the fill — a small mark only needs local context, and on real marks the cropped fill is on par with / sometimes cleaner than the full-frame fill (a tighter view gives the GAN less room to hallucinate large background structure; verified by eye on real Gemini/Doubao marks + a ground-truth reconstruction sweep). **Mask polarity is INVERTED** vs this package's 255-erase convention — the shipped ONNX wants 0=hole / 255=known, so `erase_migan` feeds `(crop_mask<=127)*255`; feeding 255=hole regenerates the whole frame into stripes (corpus-validated 2026-07, cost hours to find). ~0.19 s. This is the **preferred default fill** for the visible localize -> fill path.
-- `lama` (extra `lama`, `Carve/LaMa-ONNX` Apache-2.0, ~200 MB): `erase_lama` crops a padded region around the mask, runs at LaMa's fixed 512² input, pastes only masked pixels back. Best quality but ~4.7 GB peak — explicit opt-in only, NOT auto-selected.
-Lazy `_get_{lama,migan}_session` singletons; `{lama,migan}_available()` guard the optional imports (both == onnxruntime present). Note both extras install the same onnxruntime, so the two `*_available()` checks are identical — the fill's `auto` backend therefore resolves to MI-GAN whenever onnxruntime is present, else cv2, and big-LaMa is reachable only by an explicit `lama` backend (`--backend lama` on `erase`, or the shared fill's `backend="lama"`).
-
-**LaMa-ONNX costs ~3.5-4 GB peak RAM and ~5-6 s/call on CPU** (FFC working set, not arena — `enable_cpu_mem_arena=False` does not help), so it does NOT fit a minimal droplet; the cv2 backend (tens of MB, ~30 ms) does. LaMa quality at low RAM = serverless/GPU, mirroring how raiw.cc offloads SDXL to fal.
-
-## `invisible_watermark.py`
-
-`invisible_watermark.py` — `detect_invisible_watermark(path)` decodes the OPEN DWT-DCT watermarks (public decoder, no key) embedded by Stable Diffusion / SDXL / FLUX via the `imwatermark` library. Known fixed patterns (verified against upstream source) live in `_BITS_48` (SDXL 48-bit, FLUX.2 48-bit) and `_SD1_STRING` ("StableDiffusionV1", SD 1.x/2.x). Optional dep (extra `detect`); returns None when absent. The `detect` extra pulls **torch** transitively (invisible-watermark declares torch a hard dep, and `WatermarkDecoder` eagerly imports `rivaGan` -> `torch` at import time), so detection needs torch present even though dwtDct runs CPU-only on cv2/numpy/pywavelets — no GPU and no separate `gpu` extra required.
-
-**Unlike SynthID this is locally detectable**, but the watermark is fragile (does not survive JPEG re-encode/resize — verified gone after JPEG q90), so it confirms origin only on pristine files. Add new known patterns here. The file carries a top-of-module pyright pragma because imwatermark/cv2 ship no type stubs.
-
-## `trustmark_detector.py`
-
-`trustmark_detector.py` — `detect_trustmark(path)` decodes the OPEN, keyless **Adobe TrustMark** watermark (the soft binding behind Adobe Durable Content Credentials, `alg` `com.adobe.trustmark.P`) via the optional `trustmark` package (extra `trustmark`; pulls torch, downloads model weights on first use). Mirrors `invisible_watermark.py` (lazy singleton guarded by a double-checked `threading.Lock` so concurrent callers do not double-download the weights, top-of-module pyright pragma, returns None when absent). It detects *provenance*, not AI origin as such (TrustMark also marks human-authored content), so `identify` lists it as a watermark without setting `is_ai_generated`. Other soft-binding vendors (Digimarc/Imatag/Steg.AI/...) have no public decoder — they are only *named* via the `C2PA_SOFT_BINDINGS` scan, not decoded.
-
-**False-positive gate (added 2026-05-29):** TrustMark's `wm_present` is a BCH error-correction validity flag that spuriously validates on a content-correlated fraction of un-watermarked images — AI-generated textures trip it far more than camera photos (verified 2026-05-29 on real files: it fires on Gemini/OpenAI/Doubao output that *cannot* carry Adobe's watermark, with a random-bytes decoded secret, while signal-free camera photos did not trip it). A genuine TrustMark is a *durable* soft binding engineered to survive re-encoding, so `detect_trustmark` re-decodes after a mild JPEG round-trip (`_survives_reencode`, `_REENCODE_QUALITY` 95) and requires the same schema both times; every observed false positive collapsed (none survived even q95), so the gate is the durability property the watermark guarantees. The second decode runs only on the rare initial hit, so the cost is negligible. Do NOT remove the gate to "catch more" — a lone TrustMark hit without it is almost always content noise.
-
-## `noai/watermark_remover.py`
-
-`noai/watermark_remover.py` — the `WatermarkRemover` class has four diffusion pipelines, selected by the explicit `pipeline` ctor arg (NOT inferred from `model_id`). `sdxl`/`controlnet` share the SDXL base (`DEFAULT_MODEL_ID`); `qwen` is its own base (`QWEN_MODEL_ID`); `qwen-zimage` delegates to the fixed two-stage stack in `noai/qwen_zimage_pipeline.py`.
-
-**`sdxl`** (renamed from `default` 2026-06-09; `default` kept as a back-compat alias via `normalize_profile`) runs plain SDXL img2img (`_run_img2img`); it is the lighter opt-down alternative (no ControlNet weights).
-
-**`qwen`** (`_run_qwen`, `_load_qwen_pipeline`) runs `QwenImageImg2ImgPipeline` on `Qwen/Qwen-Image` (20B MMDiT, Apache-2.0 code AND weights). The scrub still comes from the img2img `strength`; Qwen's value is **text preservation** (incl. CJK and small text). **Metric-measured nuance (2026-06-19, `scripts/fidelity_metrics.py`, do NOT trust the eyeball here — it misled). Compare ONLY at each pipeline's oracle-confirmed scrub floor (outputs where SynthID is removed in BOTH — an equal-strength compare is invalid where it leaves one un-scrubbed; Qwen at 0.15 does not clear Gemini): Qwen wins TEXT (lower OCR CER across EN/RU/ZH, perfect Chinese) but controlnet wins FACES (higher Laplacian-variance retention and lower LPIPS — Qwen smooths faces MORE; ArcFace identity favors controlnet 0.546 vs 0.331 at the Gemini floors).** So Qwen is the better text-preserving remover, NOT a universal fidelity win — controlnet's canny edge map holds face skin detail better. Specifics: bf16 on CUDA (fp16 risks overflow on the 20B MMDiT — see the dtype branch in `__init__`); loads `QWEN_MODEL_ID` unless `--model` is overridden; the call shape lives in the pure module helper `_build_qwen_kwargs` (unit-tested without torch in `tests/test_platform.py::TestQwenKwargs`), which uses Qwen's `true_cfg_scale` (NOT SDXL's `guidance_scale` — the CLI `--guidance-scale` maps onto it; ~4.0 is typical, the SDXL default 7.5 is high for Qwen) and an explicit `negative_prompt` (`_QWEN_PROMPT`/`_QWEN_NEGATIVE`). It is CUDA/cloud-class (the 20B does not fit MPS), so `_run_qwen` has NO MPS->CPU fallback — an error propagates. `_load_qwen_pipeline` raises a clear ImportError if the installed diffusers lacks `QwenImageImg2ImgPipeline`. **CERTIFIED oracle floors (Modal A100-80GB, 2026-06-20): OpenAI 0.10 (seed-robust — clean on seeds 0-4), Gemini 0.25 (seed 0 verified on 2 images; the Gemini oracle rate-limits volume seed-repeat, so PIN a seed in prod). The Gemini floor (0.25) is HIGHER than the certified controlnet Gemini floor (0.15); `resolve_strength(..., pipeline="qwen")` carries the Qwen ladder (`_QWEN_VENDOR_STRENGTH`), so `--pipeline qwen` gets the 0.25 Gemini floor automatically -- the old manual `--strength 0.25` workaround is retired. `_build_qwen_kwargs` passes an explicit `height`/`width` from the input (floored to /16 via the pure `_qwen_target_size`); WITHOUT it the img2img pipeline defaults to a 1024x1024 SQUARE and silently squishes non-square inputs (the abba 2816x1536 case came back 1024x1024, distorting the scene and garbling text — fixed 2026-06-20, tested in `TestQwenKwargs`).** Fidelity vs controlnet was measured at the certified floors (`scripts/fidelity_metrics.py`), NOT eyeballed. **`qwen` is a MANUAL opt-in only — there is NO auto-router (one was prototyped and DROPPED, see below).** It wins ONE niche: clean body text on a plain background, NO faces (openai_1/2 CER 0.241 vs 0.385). controlnet wins FACES and **display/decorative text in a scene** (abba poster: controlnet CER 0.114 vs qwen 0.379 — canny holds letter shapes, qwen re-renders and garbles them). **`--pipeline auto` + a faces+text mixed dual-pass were built and DROPPED (2026-06-20):** on the canonical faces+text case controlnet wins EVERY metric incl. text, so grafting qwen text would only hurt; and "text→qwen" is undecidable cheaply (it is body-vs-display text that matters). The router/detector/mixed modules were removed; the geometry fix + the Qwen strength ladder were kept (they make the manual `--pipeline qwen` correct). **Do NOT retry "add a Qwen ControlNet to close the face gap" — it was built, measured, and CLOSED 2026-06-20:** a DiffSynth blockwise-canny Qwen ControlNet did not restore face skin texture (lapvar flat 0.40, canny carries edges not skin grain) and no permissively-licensed Qwen tile/detail/skin ControlNet exists anywhere (all conditioning is geometry). The Z-Image face-crop lead is now implemented as the separate `qwen-zimage` profile and has direct face metrics on two official upstream examples plus one crowded fixture. Its exact current six-output candidate is negative in the corresponding provider oracles, while broad seeded removal and text behavior remain unmeasured. Full record + the deep-research sweep in `docs/qwen-improvement-research.md`.
-
-## `noai/qwen_zimage_pipeline.py`
-
-`qwen-zimage` is the recommended high-quality, manual CUDA profile ported from `cebeuq/Synthid-Bypass` v2. `controlnet` remains the default for compatibility and cost; callers that prioritize output fidelity, especially face identity, should select `qwen-zimage`. The full-frame stage uses DiffSynth `QwenImagePipeline` with `Qwen/Qwen-Image-2512`, `lightx2v/Qwen-Image-2512-Lightning` at four steps, and `DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny`. The Lightning scheduler uses `exponential_shift_mu=log(3)`, the DiffSynth equivalent of the source graph's AuraFlow shift 3. Its default denoise is the source custom node's exact megapixel formula at adaptive level 6; an explicit `--strength` overrides that global value. Its profile seed defaults to `0`, matching the oracle-negative release candidate; an explicit seed still wins. Other profiles keep their existing random default.
-
-The face stage detects boxes on the original input with OpenCV YuNet and follows the active Impact Pack SAM path from the source graph: each box supplies both the `center-1` positive point and the box prompt; proposals at predicted IoU >= 0.93 are unioned, or the highest-IoU proposal is used when none passes; the result is intersected with the detector box. The inactive MediaPipe node in the workflow has no downstream link. YuNet uses its own calibrated score threshold, 0.5: copying the upstream YOLO threshold of 0.2 admitted background/decorative false positives and duplicate boxes, which multiplies the serial face-stage cost. The 0.5 gate retained all visible faces in the public and upstream comparison fixtures while reducing the crowded group from 36 boxes to 18 and the poster from 30 to 10. Crops expand by the source graph's factor 2.5 and run `Tongyi-MAI/Z-Image-Turbo` for eight steps. Every face uses the denoise derived from the largest face's area ratio, matching the source graph's `largest_face` mode, then pastes through the clipped SAM mask with feather 10. If SAM fails, a box-derived ellipse mask is used rather than aborting the global removal.
-
-The active graph was traced from upstream commit `3007d0351596ae0a78b7074dae7ad179710b1e48` and its linked Impact Pack implementation, not inferred from the README or node names. YuNet is one intentional substitution: the active reference path uses an Ultralytics YOLO face detector, while this package avoids adding its AGPL runtime. The first-use model download targets GitHub's media endpoint rather than the repository's 131-byte Git LFS pointer and verifies the published 232589-byte model by SHA-256 before caching it. The other runtime differences are full safetensors rather than quantized GGUF models, DiffSynth's first-order Qwen Lightning and Z-Image FlowMatch samplers rather than the graph's DPM++ 2M / SGM Uniform and `res_2s` / `bong_tangent` pairs, and the absence of the detailer's 20 px latent noise-mask feather. This port regenerates the expanded crop and composites only the feathered SAM mask; generated pixels outside the face mask are discarded. The architecture and active decision path match the graph, but the runtime is not bit-identical to ComfyUI.
-
-The implementation has its own `qwen-zimage` optional dependency group because DiffSynth, torchvision, and the additional model downloads are large. `--model` is rejected for this fixed profile. `--tile` runs only the global Qwen stage through `noai.tiling.run_tiled`; the global denoise is still derived from the full-frame megapixel count and the same seed is reused for each deterministic tile. After feather blending, YuNet, SAM, and Z-Image run once against the full original/global result, so faces are neither duplicated nor dropped at tile boundaries. The SDXL minimum-resolution floor is disabled, and CLI adaptive polish defaults off for this profile, so the two-stage result is not followed by a repository-specific post-process. DiffSynth requires the PIL input, Canny control, and explicit dimensions to agree on the same /16 latent grid: `_resize_to_target` aligns global and face pixels before `build_global_kwargs` / `build_face_kwargs`, and the global output is restored to the exact original size. Passing floored dimensions with unaligned pixels caused a real VAE/noise-grid shape mismatch on the official example 12 input; the call-shape assertions were observed failing before the fix. An explicit `--adaptive-polish` still opts in. Pure helpers cover both adaptive denoise formulas, /16 dimensions, call shapes, Canny generation, masked compositing, and the qwen-zimage tiling seam; an integration test guards dispatch from `WatermarkRemover`. A real 4096x3072, 20-tile H100 smoke completed through this exact branch on 2026-07-25 with dimensions preserved and no visible or 99th-percentile gradient outlier at a tile boundary; the measured runtime, memory, and fidelity figures are in `docs/known-limitations.md`. The exact seed-0 non-tiled release candidate is oracle-verified; tiled outputs still require their own provider-oracle check.
-
-DiffSynth normally offloads the Z-Image text encoder, DiT, and VAE to CPU after every face call. That placement dominated crowded-scene latency even though the eight diffusion steps themselves were fast. `resolve_face_model_residency` keeps the full face stack on CUDA when total VRAM is at least 64 GiB; smaller cards preserve the original offload path. Callers can explicitly override the decision through `QwenZImagePipeline.keep_face_models_on_device`. The implementation intentionally loads the stack with the normal CPU-managed config, rewrites the managed modules' offload/onload/preparing placement to their CUDA computation device, and moves them once. Loading the same models directly into CUDA cut face inference further but increased setup from 32.282 to 254.632 seconds, making a cold single request more expensive; that variant was rejected. The shipped fast-load residency changes only model placement, not weights, dtypes, prompts, seeds, schedules, masks, or compositing. On the 18-face H100 fixture it produced a pixel-identical output versus offload while reducing face regeneration from 181.764 to 38.272 seconds and total inference from 262.072 to 133.543 seconds. Setup rose only from 32.282 to 43.960 seconds, so cold setup plus inference fell from 294.354 to 177.503 seconds. Peak CUDA allocation rose from 24.364 to 43.477 GiB.
-
-Both stage prompts are constants, but DiffSynth 2.0.18 exposes their embeddings only through internal PipelineUnits and re-runs the corresponding text encoder on every call. `_cache_static_prompt_embeddings` wraps the exact prompt unit selected by its output signature and memoizes its returned tensors by prompt text. It bypasses the cache whenever `edit_image` participates, so image-conditioned embeddings cannot be reused accidentally. With CFG 1.0, the unit runner already shares the positive result with the negative branch. On a warm H100 sequence, a preceding no-face request populated the Qwen prompt cache; the 18-face case then fell from 133.543 to 78.474 seconds. The global stage fell from 89.720 to 43.334 seconds, and the face stage from 38.272 to 30.592 seconds as the fixed Z-Image prompt was encoded only once. Both the no-face and 18-face cached outputs were pixel-identical to their no-cache references, and peak allocation remained 43.477 GiB. The Qwen saving applies from the second request in a container; the Z-Image saving applies within the first multi-face request after its first face.
-
-Validation status is deliberately narrower than the existing `qwen` certification. The user reported the upstream workflow as Gemini-oracle negative. The current port has API, unit, dispatch, GPU integration runs, and a direct comparison with two official upstream before/after pairs. On 2026-07-25 the user checked all six current outputs in the provider-separated `full-clean-final-candidate-2026-07-25-by-oracle` bundle with the corresponding provider oracles and confirmed that none retained SynthID or the provider generation signal. The checked bytes used the complete `visible -> qwen-zimage -> metadata` route, the calibrated YuNet 0.5 gate, and the shipped prompt-cache/model-residency optimizations. This supersedes the earlier first-port batch check as the release-candidate result. It certifies the exact seed-0 outputs, not every seed, resolution, or content class. Broad text certification remains open, so the profile stays an experimental manual opt-in even though it is the recommended quality mode.
-
-**Direct port measurement (2026-07-24, seed 0):** on official upstream examples 10 and 12, local `qwen-zimage` retained ArcFace identity at 0.950 and 0.947 versus 0.701 and 0.548 for the current polished ControlNet output. Face LPIPS was 0.045 and 0.015 versus ControlNet's 0.105 and 0.061. The published upstream outputs retained identity at 0.976 on both and face LPIPS at 0.172 and 0.014; upstream example 10 is strongly penalized by its published downscale, while example 12 was compared at the same published dimensions. Local whole-image LPIPS / SSIM were 0.167 / 0.765 and 0.085 / 0.896, better than the published upstream 0.259 / 0.627 and 0.111 / 0.777. ControlNet still preserved more texture, but the faces drifted. The earlier crowded `gemini_3` run showed the same identity direction, 0.795 versus 0.587/0.588, while smoothing skin and changing the full frame more. The final July 25 oracle check supplies the removal verdict for the exact current candidate bytes only.
-
-Two integration failures from the first Modal passes are regression-guarded. SAM model pixels must be cast to the model's bfloat16 while geometric prompts remain float32; casting everything either fails or changes prompt semantics. SAM `pred_masks` and `iou_scores` must then be converted through float32 before NumPy because NumPy rejects bfloat16. A third visually severe failure came from accepting an unconstrained SAM mask: it split faces with hard seams. The center point plus box prompt and the final detector-box intersection are both load-bearing.
-
-**`controlnet`** (**the DEFAULT pipeline since 2026-06-09** for `invisible`/`all`/`batch` and both engine ctors; `_run_controlnet`, `_load_controlnet_pipeline`) runs `StableDiffusionXLControlNetImg2ImgPipeline` with the SDXL-native canny ControlNet `xinsir/controlnet-canny-sdxl-1.0` (`watermark_profiles.CONTROLNET_CANNY_MODEL`): the control image is `cv2.Canny(gray, 100, 200)` stacked to 3 channels (`_CANNY_LOW`/`_CANNY_HIGH`, prompt `_CONTROLNET_PROMPT` / `_CONTROLNET_NEGATIVE`).
-
-**Removal comes from the img2img regeneration (`strength`); the ControlNet only PRESERVES text and face STRUCTURE via the edge map.**
-
-No original pixels are copied or frozen, BUT **validation 2026-06-04 disproved the old "so SynthID does not survive" claim: SynthID CAN survive controlnet on photoreal/high-detail content.**
-
-At the shared low removal strength the canny edge-conditioning keeps the regeneration so close to the original that the pixel perturbation that destroys SynthID does not happen (oracle-confirmed: an OpenAI bracelet photo + a 9-face grid read **SynthID-detected** after controlnet at strength 0.10/0.15, but **SynthID-not-detected** after the `default` pipeline at the SAME strength + resolution -- only the pipeline differed).
-
-**But the reverse also holds: a flat-graphic logo/poster SURVIVED `default` while clearing controlnet** -- removal at the low strength is content×pipeline dependent and neither pipeline is universally safe; the real lever is a higher strength. See the controlnet Known-limitations bullet for the full table + root cause. Canny holds face STRUCTURE but NOT identity (the regenerated face drifts in likeness -- canny carries edges, not identity). The drifted cleaned face is the LEAST-AI state we can reach without re-introducing SynthID; the library does NOT ship a face-restore extra. Every restore approach we evaluated (GFPGAN-on-cleaned, PhotoMaker-V2 txt2img, InstantID txt2img, InstantID img2img-on-cleaned at three parameter sweeps, 2026-06-04 - 2026-06-08 Modal cert sweeps) regenerated the face from an ArcFace embedding via SDXL diffusion -- which makes the output face look MORE AI-generated, not less. Empirical conclusion in `docs/synthid-robust-identity-research-2026-06-08.md` "Empirical follow-up". For production face preservation, ship the cleaned image as-is. `controlnet_conditioning_scale` (ctor arg, default 1.0) is the structure-preservation knob. Same dtype rule as `default` (fp32 on cpu/mps, fp16 only on cuda/xpu; the fp16-fixed SDXL VAE `_SDXL_FP16_VAE_ID` is swapped in on fp16 GPUs -- issue #29) and the same MPS->CPU fallback (reload on cpu/fp32, drop a non-cpu generator, retry once).
-
-**Tiled diffusion (`tile`/`tile_size`/`tile_overlap` ctor-path args, CLI `--tile`, issue #10):** for large inputs that OOM at native resolution, `remove_watermark` can process the diffusion pass in overlapping sliding-window tiles instead of one forward pass — the lossless alternative to a `--max-resolution` downscale. For SDXL, ControlNet, and base Qwen, the single-image generation closure was refactored into `_generate_one(img)` (dispatches controlnet/img2img, generator shared so the seed advances deterministically across tiles), and `_generate()` routes it through `noai.tiling.run_tiled` when `tile` is set AND `max(init_image.size) > tile_size` (a sub-tile image runs one pass unchanged). The ControlNet canny edge map is rebuilt per tile inside `_generate_one`, so structure preservation is tile-local. `qwen-zimage` takes a different route: `_generate()` dispatches once to `QwenZImagePipeline.run`, that runtime tiles only `_run_global`, then performs one full-frame face stage after blending. See `noai/tiling.py` below and the tiled-diffusion subsection in `docs/known-limitations.md` for the geometry, the partition-of-unity blend, and the quality caveat.
-
-## `noai/tiling.py`
-
-Pure sliding-window tiling for the diffusion path (no torch import; numpy/PIL only). `plan_tiles(w, h, tile_size, overlap)` returns a row-major grid of uniform-size `Tile` boxes — every tile is exactly `tile_size`, with the last tile on each axis pulled back flush to the far edge (`_axis_positions` clamps a pathological `overlap >= tile` to `tile - 1` so the step stays >= 1). `feather_weights(w, h, overlap)` is a separable linear taper (1 in the interior, ramping toward each edge) floored at `_WEIGHT_EPS` so it is **strictly positive everywhere** — that makes the normalized `accum / weight_sum` blend a partition of unity, so identical/unchanged tiles reconstruct the input exactly (the seam-free guarantee). `run_tiled(generate_tile, image, tile_size, overlap, set_progress)` is the orchestration loop: crop each planned tile, call `generate_tile` (one diffusion pass on a single PIL tile — injected, so this stays decoupled from the pipeline), resize a latent-grid-rounded result back to the exact tile size, and feather-accumulate. All three are unit-tested without the model (`tests/test_tiling.py`: axis math, grid coverage, taper shape/symmetry/positivity, identity reconstruction, per-tile call count, and the resize-back path). New blend tuning belongs in these pure helpers, not inlined into the runner.
-
-`feather_region_composite(base, regenerated, box, *, feather)` is the pure region-targeted compositor for **AI-enhanced composites** (roadmap P1#8; `identify` `ai_source_kind == "enhanced"`, digitalSourceType `compositeWithTrainedAlgorithmicMedia`). It blends `regenerated` over `base` inside `box = (x, y, w, h)` with a separable linear taper of `feather` px at the box edges (the taper anchors to ~0 at the boundary, so unlike `feather_weights` it is NOT floored — the result equals `base` EXACTLY outside the box), preserving dtype and supporting HxW or HxWxC. It backs `WatermarkRemover.remove_watermark(region=..., region_feather=...)`: the remover regenerates the frame (or tiles), then composites only the AI box back over the original input, so the real photo outside the box stays pixel-exact and only the AI region is scrubbed. The box is caller-supplied (a C2PA composite manifest carries no reliable machine-readable region); the no-model lossless region path remains `region_eraser.erase`. Unit-tested in `tests/test_tiling.py::TestFeatherRegionComposite` (outside-box exactness, interior == regenerated, hard-paste at feather 0, monotonic seam ramp, dtype/grayscale/clamp/empty-box/shape-mismatch).
-
-## `auto_config.py` (REMOVED 2026-06-09)
-
-**`auto_config.py` + the content-detection layer were REMOVED 2026-06-09.**
-
-History: `auto_config.plan()` was a content-adaptive planner that detected faces/text/edges (bundled OpenCV YuNet + PP-OCRv3 DBNet models) to route the pipeline and toggle the adaptive polish. Once `controlnet` became the default-and-only auto pipeline (it no longer downgrades a structure-less image to `sdxl`) and the adaptive polish was confirmed to **self-gate by detail level** (`humanizer.adaptive_polish` no-ops when the cleaned image already meets the input's Laplacian variance, so it does real work only on over-smoothed photo/face texture and ~nothing on text/flat), the detection no longer changed any behavior — it only annotated a `reason` string. So the whole layer was deleted: `auto_config.py`, `tests/test_auto_config.py`, and the two detection assets (`assets/face_detection_yunet_2023mar.onnx`, `assets/text_detection_ppocrv3_2023may.onnx`, ~2.6 MB).
-
-**`--auto` is now a DEPRECATED no-op** (`cli._resolve_auto_polish`): controlnet is already the default pipeline AND the adaptive polish is ON by default, so `--auto` has nothing left to do — it only prints a deprecation warning and passes `adaptive_polish` through unchanged (an explicit `--no-adaptive-polish` still wins). (Originally it re-enabled the polish; once the polish default flipped to ON the same day, the parameter-source branch became dead and was dropped.) The **adaptive polish itself lives on** in `humanizer.adaptive_polish` (CLI `--adaptive-polish/--no-adaptive-polish`, **ON by default since 2026-06-09 for the original profiles** — it self-gates to a no-op where there is no detail deficit; `qwen-zimage` defaults it off to preserve its upstream-matching output, and an explicit flag overrides either default) — see the `humanizer` test note. `batch` resolves the polish once before the loop (one warning) and caches the invisible engine per pipeline (`ctx.obj["_inv_engines"]`).
-
-## Content `--pipeline auto` router + faces+text mixed dual-pass — PROTOTYPED and DROPPED (2026-06-20)
-
-A `--pipeline auto` content router (`pipeline_router.py` + `content_detect.py`: Haar faces + MSER text → route text→qwen / faces→controlnet / both→mixed) and a faces+text **mixed dual-pass** (`mixed_pipeline.py`: scrub the whole frame on BOTH pipelines, then graft the qwen text regions onto the controlnet base via `tiling.feather_region_composite`) were built, run on Modal (the abba poster: faces + display text), measured, and **removed**. Why it failed:
-- On the canonical faces+text image **controlnet wins EVERY metric, including text** (CER 0.114 vs qwen 0.379; ID 0.64 vs 0.36; lapvar 0.71 vs 0.59) — canny holds the existing letter shapes, qwen re-renders display/decorative text and garbles it. So grafting qwen text onto the controlnet base only HURTS.
-- qwen beats controlnet on text ONLY for clean body text on a plain background with no faces (openai_1/2) — a niche where there are no faces to route around anyway, so `--pipeline qwen` alone covers it. The faces+clean-body-text intersection is near-empty.
-- "text→qwen" is not cheaply decidable: it is body-vs-display text that matters, which face/text detectors can't tell apart. MSER also over-fired (47% of the busy poster, incl. faces).
-
-KEPT from that work (independently valid for the manual `--pipeline qwen`): the qwen **geometry fix** (`_qwen_target_size` + `_build_qwen_kwargs` height/width — qwen squished non-square inputs to 1024² without it) and the **pipeline-aware `resolve_strength`** Qwen ladder (Gemini 0.25). Also kept: the `fidelity_metrics.py` one-to-one face matcher. The throwaway Modal eval scripts were removed after the run (findings recorded here and in `docs/qwen-improvement-research.md`).
-
-## `upscaler.py`
-
-`upscaler.py` — optional Real-ESRGAN pre-diffusion super-resolution for small inputs (spandrel boundary, top-of-file pyright pragma). `is_available()` gates on spandrel+torch (via `importlib.util.find_spec`); `upscale(bgr, device=None)` loads a lazily-built spandrel `ImageModelDescriptor` singleton (double-checked lock) and upscales by the model's native factor (x2), with a non-CPU→CPU device fallback mirroring the diffusion engine's MPS→CPU retry. Weights (`RealESRGAN_x2plus.pth`, BSD-3-Clause) download on first use to the `torch.hub` checkpoints cache; never bundled. Used only when UPscaling to the `min_resolution` floor (a `max_resolution` downscale always uses Lanczos). The wiring is `InvisibleEngine._esrgan_upscale(pil, target)` — Real-ESRGAN at native factor, then a Lanczos resize to the exact target, falling back to a plain Lanczos resize if the extra is absent or the model errors (so an optional upscaler can never break removal). The default `--upscaler` is `lanczos` (cv2, no deps).
-
-**ESRGAN is a generic photo/texture GAN with no face/glyph prior**, so it best fits photo/texture content and can degrade faces (glassy/asymmetric eyes -- the diffusion pass regenerates faces so the full-pipeline final recovers) and thin/small text (the GAN invents wrong strokes, and low-strength diffusion will not fix it). Verified 2026-06-04: isolated upscale lap-var ~5x Lanczos on faces+textures but glassy eyes; end-to-end `invisible` final lap-var 1634 vs Lanczos 663 with natural faces (diffusion cleaned the artifact). Kept a **manual opt-in knob** (the auto plan never selects it) with `lanczos` the default; not content-gated by design (use Lanczos for text-heavy inputs). spandrel is MIT and pulls no basicsr. Unit-tested without the model: `tests/test_upscaler.py` (availability guard + the not-installed RuntimeError) and `tests/test_invisible_engine.py::TestEsrganUpscale` (the three `_esrgan_upscale` branches via a monkeypatched `upscaler`).
-
-## `image_io.py`
-
-`image_io.py` — Unicode-safe cv2 IO (issue #17). `imread(path, flags=None)` / `imwrite(path, img)` wrap `np.fromfile`+`cv2.imdecode` / `cv2.imencode`+`tofile` so non-ASCII paths work on Windows -- bare `cv2.imread`/`cv2.imwrite` use the platform ANSI code-page API there and fail (empty decode + `can't open/read file`) on Chinese/Cyrillic/accented filenames. `imread` keeps `cv2.imread` semantics (defaults to `IMREAD_COLOR`, returns `None` on missing/empty/undecodable).
-
-**Every cv2 file read/write in the package routes through here; do not call `cv2.imread`/`cv2.imwrite` directly.**
-
-`imwrite` returns `False` on an unwritable path (`OSError` caught) instead of raising, matching `cv2.imwrite` semantics. macOS/Linux already accept UTF-8 paths, so it is behavior-neutral there (the bug only reproduces on Windows).
-
-**`to_bgr(image)` (added 2026-06-09)** is the shared channel normalizer: promotes 2D grayscale / (h,w,1) / 4-channel BGRA to 3-channel BGR (a 3-channel input is returned unchanged, no copy). Use it instead of inlining the `cvtColor(GRAY2BGR/BGRA2BGR)` branch — the gemini engine and the `TextMarkEngine` base both route through it so a grayscale/BGRA input (a real Gemini-app export is opaque RGBA) does not crash the `axis=2` channel reductions. cv2/numpy are imported lazily inside the functions, so the module is cheap to import in a bare env.
-
-## CLI commands (`cli.py`)
-
-Full per-command behavior for the skip/exit branches summarized in `CLAUDE.md`'s "How to run". The CLI distinguishes three exit codes: success (0), hard error (1), and a "nothing to do" code (2, `EXIT_NO_VISIBLE_MARK` / `EXIT_NO_INVISIBLE_SIGNAL`) so a wrapping service (raiw.cc) can surface guidance instead of treating an unchanged image as done (the production "it didn't work" / score-0 trap).
-
-**Every single-image command's `source` argument declares `dir_okay=False`** (2026-07-20). `click.Path(exists=True)` accepts a directory unless told otherwise, so `identify <dir>` sailed past argument parsing and raised `IsADirectoryError` out of `metadata.scan_head`'s `open()` — a traceback, not a usage error. Refusing it at the argument layer is the right place: every command gets it, and none needs its own check. (`batch`'s `directory` argument was already correct with `file_okay=False`.) Found by the Tier E adversarial sweep; regression: `tests/test_cli_robustness.py::TestDirectoryInputIsRejected`.
-
-### `all`
-
-Full pipeline (visible + invisible + metadata). Same diffusion knobs as `invisible`, plus the visible-pass `--backend auto|cv2|migan|lama` (default `auto`) that picks the fill for the localize -> fill visible removal. **When the `[gpu]` extra is absent, step 2 (invisible/SynthID) is skipped** — `all` still writes an output (visible mark + metadata stripped) but prints a prominent end-of-run banner ("the invisible (SynthID) watermark was NOT removed") AND exits **non-zero** (1), so a skipped SynthID pass is not mistaken for a clean result (the recurring #14/#47 trap, where the old quiet inline warning was missed). `invisible` already hard-errors without the extra; only `all` continued, hence the loud end-banner. Regression-guarded by `tests/test_cli.py::TestAllCommand::test_all_loud_warning_and_nonzero_exit_when_gpu_missing`. **No-signal skip (P0#5):** step 2 also runs the same `has_invisible_target` gate (see `invisible` below) — when no invisible watermark is detectable and `--force` is not set, step 2 is skipped and the pixels are left intact, but unlike the GPU-missing skip this is a **SUCCESS (exit 0)**: the visible pass + metadata strip still ran and a file is written (the message says so without claiming the image is clean). Distinct exit semantics by design: GPU-missing = couldn't do the work (non-zero); no-signal = nothing to do (zero). Regression-guarded by `test_all_skips_invisible_on_no_signal_but_succeeds`. **Test trap:** any `all` test that exercises the full pipeline MUST `patch("remove_ai_watermarks.invisible_engine.is_available", return_value=True)` — CI installs core+dev only (no `[gpu]`), so an unpatched `all` test takes the skip branch and now hits the non-zero exit. This passed locally (gpu present → `is_available()` True) but red-failed every matrix cell on the v0.11.0 commit (`test_all_basic`/`test_all_visible_step_uses_registry` asserted exit 0); both now patch `is_available` True.
-
-### `invisible`
-
-Diffusion SynthID removal. The `--tile/--no-tile` knob is the *lossless* alternative to a `--max-resolution` downscale for large inputs that OOM on MPS/GPU: it engages only when the long side exceeds `--tile-size` (default 1024); tiles are feather-blended over `--tile-overlap` px (default 128); pair with `--max-resolution 0`. `--cpu-offload/--no-cpu-offload` trades speed for lower CUDA VRAM use: SDXL, ControlNet, and base Qwen call Diffusers `enable_model_cpu_offload(device="cuda")`, which moves whole model components between CPU and GPU; `qwen-zimage` instead forces its face stack to use the existing offload path rather than automatic high-VRAM residency. The flag has no effect on CPU/MPS and fails loudly if a CUDA Diffusers pipeline lacks the offload method. `--adaptive-polish` is a detail-targeted polish that self-gates to a no-op where there is no deficit; it defaults off only on `qwen-zimage`. `--auto` is deprecated and now a no-op that only warns. **No-signal skip (P0#5, roadmap):** before the diffusion runs, the command checks `identify.has_invisible_target(source)` (the `ProvenanceReport.ai_from_metadata` union: C2PA AI-issuer / SynthID proxy, IPTC, AIGC, local gen params, EXIF/xAI, open DWT-DCT / TrustMark — visible marks do NOT count, they are a separate pass). When nothing is locally detectable it does NOT regenerate (that would only degrade a clean image — the dominant paid score-0 cause on no-watermark uploads): it writes NO output, prints guidance that does NOT claim the image is clean (a pixel SynthID is undetectable once its metadata proxy is gone), and exits **`EXIT_NO_INVISIBLE_SIGNAL` (2)** — same value/role as the visible `EXIT_NO_VISIBLE_MARK`. `--force/--no-force` (**default skip = ON**) runs the scrub regardless. The check fails SAFE (a detector exception → run, since leaving a watermark on a paid removal is worse than over-regenerating). Helpers `cli._no_invisible_signal_exit` + `identify.has_invisible_target`; regression-guarded by `tests/test_cli.py::TestInvisibleCommand::{test_invisible_no_signal_skips_and_exits_two,test_invisible_force_runs_scrub_on_no_signal,test_invisible_runs_without_force_when_signal_present,test_invisible_cpu_offload_flows_to_engine}`, `tests/test_cli.py::TestAllCommand::test_all_cpu_offload_flows_to_engine`, `tests/test_cli.py::TestBatchCommand::test_batch_cpu_offload_flows_to_cached_engine`, and `tests/test_identify.py::TestHasInvisibleTargetFailSafe`. **Test trap:** any `invisible`/`all`/`batch` test that exercises the diffusion path on a signal-LESS fixture (e.g. the synthetic `sample_png`) MUST pass `--force`, or the new gate skips step 2 (so `mock_engine.remove_watermark` is never called / `invisible` exits 2).
-
-### `visible`
-
-Known-visible-mark removal by **localize -> fill**: each detected mark is localized to a binary full-frame footprint mask, then one shared, swappable fill inpaints that mask. `--backend auto|cv2|migan|lama` (default `auto`) picks the fill: `cv2` (classical inpaint, no deps, the floor), `migan` (MI-GAN ONNX, the memory-tight pick where LaMa will not fit), `lama` (big-LaMa ONNX, best quality, heavier, auto-preferred when a learned backend is available); `auto` = LaMa > MI-GAN > cv2, best available (LaMa is auto-preferred when a learned backend is present; a memory-tight deploy pins migan). `--sensitivity auto|strict|assume-ai` (default `auto`) controls how hard a borderline mark is trusted (see the registry section: the visual detectors are metadata-independent; `auto` relaxes a mark only on same-product evidence, `assume-ai` relaxes every mark on the caller's AI assertion, subject to the assumed-trust confidence floor where the vendor is unconfirmed — the only path to higher recall on a metadata-stripped screenshot). `--backend` and `--sensitivity` are shared across `visible`/`all`/`batch`. Detection keys on each mark's own shape, and under `auto` the trust gate is relaxed when local metadata confirms the vendor (a Google/Gemini C2PA issuer relaxes gemini, a China-AIGC label relaxes doubao/jimeng, `samsung_genai` relaxes samsung), so a moved or re-rendered mark is still caught. `--mark auto` (default) removes EVERY detected mark in one pass (`registry.remove_auto_marks`, not the single strongest -- a Jimeng-basic image carries both the top-left pill and the bottom-right wordmark) from: the Gemini sparkle, the Doubao "豆包AI生成" text strip, the Jimeng "★ 即梦AI" wordmark, the Qwen "千问AI生成" text strip, the Samsung Galaxy AI "✦ Contenuti generati dall'AI" strip (bottom-LEFT, Italian-locale detection), and the capture-less Jimeng "AI生成" pill (top-left, `pill_engine`). The pill's weak edge-NCC detector is gated in `remove_auto_marks` via `_keep_pill` (32k real-upload corpus validation 2026-07): never on Doubao or Qwen, and two confirmation arms since metadata confirms the platform, not pill presence. (1) The bottom-right wordmark fired — ~94% precise and survives metadata-STRIPPED uploads (screenshots / re-saves) — removes the pill unrestricted. (2) TC260 metadata confirms Jimeng (`"jimeng" in provenance`, from `cli._visible_provenance`) OR the caller asserts AI (`sensitivity == "assume_ai"`), no wordmark — **re-measured 2026-07-18 on 149 blind-labelled pill fires: 21% precise raw (CI 16-29%), 29% (CI 20-40%) among the flat footprints the guard actually PASSES, 14% among those it blocks** — its false fires are textured ceilings/walls that the fill visibly SMEARS — removes the pill ONLY when the top-left footprint is flat enough for an invisible fill (`pill_engine.footprint_is_flat`, median-Sobel ≤ `_FLAT_TEXTURE_MAX`; the flatness guard holds even under `assume_ai`). No confirmation → never removed. `--mark gemini|doubao|jimeng|qwen|samsung|jimeng_pill` forces one (choices come from the registry). Corpus validation: doubao and jimeng localize + remove at ~100% with clean footprints (the filled region blends into its surroundings within a few LAB levels, no color shift, no dark pit); clean images with no vendor signature had 0% false removal. For arbitrary logos/objects use `erase`. **When `--mark auto` finds no known mark (the common case — ~74% of real uploads carry no registered visible mark), the command does NOT silently re-serve the input as a finished result.** It runs a cheap metadata-only `identify`, prints actionable guidance (if the image carries an invisible/metadata mark, e.g. an OpenAI/Gemini C2PA image, it points to `all`; otherwise it does NOT imply the image is clean -- it warns that an invisible pixel watermark like SynthID cannot be detected once the metadata proxy is gone and routes to both `all` and `erase --region`), writes NO output file, and exits **`EXIT_NO_VISIBLE_MARK` (2)** — distinct from success (0) and a hard error (1) so a wrapping service (raiw.cc) can surface the message instead of treating the unchanged image as done (the production "it didn't work" / score-0 trap). Same handling for an explicit `--mark <name>` that is not detected. Helper `cli._no_visible_mark_exit`; regression-guarded by `tests/test_cli.py::TestVisibleCommand::test_visible_auto_no_mark_exits_two_with_eraser_hint` and `test_visible_auto_no_mark_routes_to_all_when_metadata`. `--no-detect` still forces the gemini fallback and proceeds (exit 0).
-
-### `batch`
-
-Process every supported image in a directory (output defaults to `<directory>_clean/`, set with `-o`). `--mode visible|invisible|metadata|all` (default `visible`); the invisible/all path reuses the **full `invisible` knob set** (`--strength`/`--steps`/`--guidance-scale`/`--pipeline`/`--controlnet-scale`/`--model`/`--device`/`--max-resolution`/`--min-resolution`/`--upscaler`/`--seed`/`--hf-token`/`--humanize`/`--unsharp`/`--adaptive-polish`/`--tile`/`--tile-size`/`--tile-overlap`/`--cpu-offload`/`--force`), plus `--backend` for the visible localize -> fill pass. `--adaptive-polish` is ON by default except on `qwen-zimage`; `--auto` is deprecated and a no-op that only warns. **No-signal skip (P0#5):** in invisible/all mode each image runs the same `has_invisible_target` gate — a signal-less image is skipped (no diffusion); in `invisible` mode the input is copied through to the output dir so it stays complete, in `all` mode the visible-removed result is kept and metadata is still stripped. `--force` scrubs every image regardless. One engine cached per pipeline; the polish is resolved once before the loop. **Exit code (`batch` used to always exit 0, hiding failures):** `cmd_batch` raises `SystemExit(1)` when any image errored, OR when a `--mode invisible`/`all` image carried an invisible signal but the GPU extra was absent so its SynthID scrub was skipped — mirroring single `all`, it emits a loud "the invisible watermark was NOT removed on N image(s)" warning and (invisible mode) copies the input through so the output dir stays complete, rather than silently dropping the signal-bearing files that most needed processing. `_process_batch_image` returns that skipped-scrub flag; the loop tallies it. Regression-guarded by `tests/test_cli.py::TestBatchCommand::{test_batch_errors_exit_nonzero, test_batch_invisible_gpu_missing_writes_output_and_exits_nonzero}`.
+1. keep model-free logic in pure helpers where possible;
+2. test option propagation and dispatch without downloading models;
+3. run a real model smoke for the changed model path;
+4. treat provider-verifier results as specific to the exact checked output;
+5. update [known limitations](known-limitations.md).

docs/python-api.md

@@ -0,0 +1,173 @@
+# Python API
+
+Use the high level API for normal application integration. Low level detector
+and pipeline modules are intended for maintainers and specialized workflows.
+
+## Remove visible marks
+
+```python
+import remove_ai_watermarks as raiw
+
+result, removed = raiw.remove_visible(
+    "watermarked.png",
+    "clean.png",
+)
+```
+
+The function returns:
+
+- the result as a BGR NumPy array;
+- a list of labels that were removed.
+
+An empty `removed` list means that no registered visible mark was selected. It
+does not prove the image has no metadata or invisible watermark.
+
+### Path input
+
+For a path input, `remove_visible`:
+
+- reads metadata provenance for the default `auto` sensitivity;
+- preserves a separate alpha channel;
+- writes the output when an output path is supplied;
+- strips AI metadata from the written output by default;
+- preserves the original bytes for a same-format no-op copy.
+
+```python
+result, removed = raiw.remove_visible(
+    "watermarked.png",
+    "clean.png",
+    sensitivity="auto",
+    backend="auto",
+    strip_metadata=True,
+)
+```
+
+Set `write_noop=False` if the output path must remain untouched when nothing is
+removed:
+
+```python
+result, removed = raiw.remove_visible(
+    "input.png",
+    "clean.png",
+    write_noop=False,
+)
+```
+
+### Array input
+
+Array inputs are BGR NumPy arrays. They do not carry file metadata or a separate
+alpha plane:
+
+```python
+import cv2
+import remove_ai_watermarks as raiw
+
+image = cv2.imread("input.png")
+result, removed = raiw.remove_visible(image, backend="cv2")
+```
+
+## Inspect provenance
+
+Get the vendor keys used by visible removal:
+
+```python
+import remove_ai_watermarks as raiw
+
+vendors = raiw.visible_provenance("input.png")
+```
+
+Get the full provenance report:
+
+```python
+from pathlib import Path
+
+from remove_ai_watermarks.identify import identify
+
+report = identify(Path("input.png"))
+print(report.platform)
+print(report.signals)
+```
+
+Use `check_visible=False` and `check_invisible=False` for metadata only
+inspection:
+
+```python
+report = identify(
+    Path("input.png"),
+    check_visible=False,
+    check_invisible=False,
+)
+```
+
+## Strip metadata
+
+```python
+from pathlib import Path
+
+from remove_ai_watermarks.metadata import has_ai_metadata, strip_and_verify
+
+source = Path("input.png")
+output = Path("clean.png")
+
+if has_ai_metadata(source):
+    output_path, surviving_markers = strip_and_verify(source, output)
+    if surviving_markers:
+        raise RuntimeError(
+            f"AI metadata remains in {output_path}: {surviving_markers}"
+        )
+```
+
+Use `strip_and_verify` when your application reports that stripping succeeded.
+It checks the written output and returns `(output_path, surviving_markers)`.
+Treat a nonempty `surviving_markers` mapping as a failure.
+
+`remove_ai_metadata` is the lower level fail-safe transformer. It may copy an
+undecodable input through unchanged, so its return alone must not be presented
+as proof that metadata was removed.
+
+## Remove invisible watermarks
+
+```python
+from pathlib import Path
+
+from remove_ai_watermarks.invisible_engine import InvisibleEngine
+
+engine = InvisibleEngine(
+    pipeline="controlnet",
+    device=None,
+    cpu_offload=False,
+)
+
+engine.remove_watermark(
+    Path("watermarked.png"),
+    Path("clean.png"),
+)
+```
+
+`device=None` selects the device automatically. Supported explicit values are
+defined by the CLI and runtime device resolver.
+
+For limited CUDA memory:
+
+```python
+engine = InvisibleEngine(
+    pipeline="controlnet",
+    cpu_offload=True,
+)
+```
+
+For the CUDA only high fidelity profile:
+
+```python
+engine = InvisibleEngine(pipeline="qwen-zimage")
+```
+
+The `qwen-zimage` extra must be installed for that profile.
+
+The full `remove_watermark` signature includes strength, steps, guidance,
+seeding, tiling, resolution, upscaling, and postprocessing controls. Read the
+method signature in
+[`invisible_engine.py`](../src/remove_ai_watermarks/invisible_engine.py) or use
+the CLI guide for the concepts.
+Defaults can differ between the Python method and CLI profile resolution, so
+pass values explicitly when reproducibility matters.

docs/qwen-improvement-research.md

@@ -1,5 +1,9 @@
 # Qwen-Image improvement research (2026-06-20)
 
+> Research archive. This page records experiments and decisions from the date
+> above. It may mention prototypes or defaults that were later changed. Use the
+> user guides and current source code for the supported interface.
+
 Cited research behind the decision **"ship the `qwen` pipeline as-is, or improve it
 first?"** Produced by the multi-source deep-research harness (5 search angles, 22
 sources fetched, 85 claims extracted, 25 verified by a 3-vote adversarial check, 20

docs/release-and-distribution.md

@@ -1,39 +1,85 @@
 # Release and distribution
 
-> Relocated verbatim from `CLAUDE.md` on 2026-06-11 to keep the always-loaded
-> context small. Long single-line entries were reformatted into paragraphs;
-> no content was changed or summarized.
+This page describes the release behavior defined in this repository. External
+registry state can change independently, so verify it during a release.
 
-Release flow and every distribution channel (PyPI, Homebrew tap, conda-forge,
-ComfyUI Registry, HF Space), plus sdist/build-backend history. The CI summary
-stays in `CLAUDE.md`; read this before cutting a release.
+## Release sources of truth
 
-`publish.yml` stays release-only and now verifies the release tag matches the `pyproject.toml` version (fails the build on a mismatch) before building, then uploads via `uv publish` (PyPI trusted publishing over OIDC, no token — replaced the `pypa/gh-action-pypi-publish` action so the upload no longer depends on that action's bundled twine accepting the Metadata-Version; the `id-token: write` permission + `pypi` environment + workflow filename are unchanged, so PyPI's trusted-publisher entry still matches).
+The package version appears in:
 
-**Release flow:** bump the version in `pyproject.toml` + `src/remove_ai_watermarks/__init__.py` + `uv.lock` (the project's own `[[package]]` entry — find it with `grep -n 'name = "remove-ai-watermarks"' uv.lock`, the `version =` line right below it, ~line 2246), commit `chore(release): vX.Y.Z`, `git tag -a vX.Y.Z -m vX.Y.Z` (annotated — `git tag` without `-m` errors here), push `main` + the tag, then `gh release create vX.Y.Z` — **PyPI publish triggers on the GitHub Release `published` event, NOT on the tag push**, so the tag alone does not publish.
+- `pyproject.toml`;
+- `src/remove_ai_watermarks/__init__.py`;
+- the root package entry generated in `uv.lock`.
 
-**After the PyPI sdist is live, bump the Homebrew formula** in the separate public tap repo `wiltodelta/homebrew-tap` (`Formula/remove-ai-watermarks.rb`): update `url` to the new sdist URL (from `https://pypi.org/pypi/remove-ai-watermarks/<version>/json`, the `sdist` entry's `url`) and `sha256` to its hash, commit + push there — otherwise `brew install wiltodelta/tap/remove-ai-watermarks` keeps installing the old version. The formula is a core-only venv that pip-installs the sdist (no vendored resources, so pip pulls the binary numpy/opencv wheels per platform at install time); only those two lines change per release.
+Update the first two, then refresh the lock file with uv. Do not edit a
+line-number-specific location in `uv.lock`; its package order changes.
 
-**This is now AUTOMATED:** the main repo's `.github/workflows/distribute.yml` fires on the GitHub Release `published` event, waits for the sdist to appear on PyPI (poll loop, the Release event races publish.yml's upload), rewrites the formula's `url`+`sha256`, and pushes to the tap using the `HOMEBREW_TAP_TOKEN` repo secret (a fine-grained PAT with Contents:write on `homebrew-tap`). The SAME workflow also factory-rebuilds the HF Space (`HfApi.restart_space(..., factory_reboot=True)`, `HF_TOKEN` secret) so the Space reinstalls the new sdist (it pins `remove-ai-watermarks>=...` and only re-resolves on a rebuild). The manual Homebrew steps above are the fallback / what the workflow automates — a normal release needs no Homebrew or HF action.
+## Publish flow
 
-**Where the HF Space's own source lives (its demo code, NOT this library):** the private repo **`wiltodelta/raiw-hf-space`** (locally `~/Documents/GitHub/raiw-hf-space`) — `app.py` (CPU-core Gradio demo), `requirements.txt`, `README.md` (the Space card), `assets/`, `examples/`. **Deploying it is a plain `git push` to that repo's `main`:** its `.github/workflows/sync-to-hf.yml` mirrors the files onto the Space via the Hub API (`HfApi.upload_folder`, secret `HF_TOKEN` = a **write**-role token), which adds a commit on top of the Space's own history — deliberately NOT a `git push --force` to the Space, which would clobber it. Do NOT edit the Space through the huggingface.co web UI any more: that was the pre-2026-07-16 workflow (it is why every Space commit before then is authored `@users.noreply.huggingface.co` and why no local write token ever existed), and a web edit now silently diverges from the GitHub source of truth. Note the demo tracks the library's **CPU-core** surface only (`identify` / `visible` / `metadata`); the invisible/SynthID path needs a GPU and stays out. Two distinct automations touch the Space and must not be confused: `sync-to-hf.yml` (in `raiw-hf-space`) ships **demo code changes**; `distribute.yml` (in this repo) factory-rebuilds the Space on a **library release** so its `remove-ai-watermarks>=...` pin re-resolves to the new version.
+PyPI publishing is triggered by a published GitHub Release, not by a tag push
+alone.
 
-**If the distribute.yml Homebrew job fails with "Bad credentials" (or the tap push 403s),** the `HOMEBREW_TAP_TOKEN` secret has expired or been revoked — fine-grained PATs expire on a fixed date, so this recurs. Fix: rotate the PAT (a fine-grained token with Contents:write on `wiltodelta/homebrew-tap`), update the `HOMEBREW_TAP_TOKEN` repo secret, then re-run the failed job (`gh run rerun <run-id> --failed`). While the token is being rotated, the manual formula bump above unblocks the release. The same rotate-secret-and-rerun applies to any distribute.yml credential failure (`HF_TOKEN` for the Space rebuild).
+The expected sequence:
 
-**Other distribution channels:** (1) **conda-forge** — recipe source of truth committed at `packaging/conda/recipe.yaml` (v1 `recipe.yaml`, noarch core-only: pillow/piexif/numpy/py-opencv/click/python-dotenv); the initial submission is `conda-forge/staged-recipes` PR #33674 (went green only after **`pip_check: false`** in the python test — rattler-build's `pip check` defaults to ON and fails on the ancient conda-forge `piexif py_2` build's stale metadata with "piexif 1.1.3 is not supported on this platform", though the package installs/imports/works; keep it disabled). Once that merges and the `remove-ai-watermarks-feedstock` exists, the `regro-cf-autotick-bot` auto-opens a version-bump PR on the feedstock when each new PyPI sdist is detected — just review + merge it (hand-edit only if run-deps changed; keep `packaging/conda/recipe.yaml` in sync as the reference copy). (2) **ComfyUI Registry** — the node package is a SEPARATE repo `wiltodelta/ComfyUI-remove-ai-watermarks` with its OWN `pyproject.toml` `version` (independent of the library version). Publish a new node version by bumping that `version` in the node repo's `pyproject.toml` and pushing to `main` — the node repo's `.github/workflows/publish.yml` (`Comfy-Org/publish-node-action@main`, triggered on a push that touches `pyproject.toml`, secret `COMFY_REGISTRY_TOKEN`) **auto-publishes** it; `comfy node publish --token <registry-PAT>` is the manual/local fallback. It is NOT auto-published on a library release (the node has its own version), so only bump it when the node code or its `remove-ai-watermarks>=` dependency floor changes.
+1. update the version sources and lock file;
+2. run the complete project gate;
+3. commit the release change;
+4. create an annotated `vX.Y.Z` tag;
+5. push the commit and tag;
+6. publish the GitHub Release.
 
-**Sdist must exclude `data/`** (`[tool.hatch.build.targets.sdist] exclude = ["/data"]`): hatchling's default sdist bundles all VCS-tracked files, so the committed `data/` test corpora (the multi-hundred-MB synthid_corpus images + the visible-mark captures) pushed the **0.8.0** sdist past PyPI's per-project file-size limit (400 "File too large") — the wheel uploaded but the sdist was rejected, so 0.8.0 shipped wheel-only and 0.8.1 carried the fix. The wheel only ships `src/` (via `[tool.hatch.build.targets.wheel] packages`), so it was never affected.
+`.github/workflows/publish.yml` then:
 
-**A failed PyPI upload of one artifact still leaves the other live and you cannot re-upload the same version** — fix the build and cut the next patch.
+1. checks that the release tag matches `pyproject.toml`;
+2. builds the package with uv;
+3. publishes with `uv publish` through PyPI trusted publishing.
 
-**Build backend is unpinned `hatchling`** (`[build-system] requires`) since 2026-06-09. History: it was pinned `<1.31` because hatchling 1.30.0 made Metadata-Version 2.5 (PEP 794) the default and the twine bundled in `pypa/gh-action-pypi-publish@release/v1` rejected it (`"'2.5' is not a valid Metadata-Version"`), which **failed the v0.8.3 PyPI upload on 2026-06-01**; hatchling 1.30.1 reverted the default to 2.4. After the workflow moved to `uv publish` (whose uploader accepts 2.5) the pin was belt-and-suspenders only, and once v0.9.0 + v0.10.0 both published wheel+sdist through that path (verified on PyPI) it was dropped. If a future hatchling flips the default to 2.5 again and some consumer chokes, re-pin with a dated comment.
+The workflow uses GitHub OIDC through the `pypi` environment. It does not read a
+PyPI API token from the repository.
 
-## Dependency CVE-resolution history (`uv-secure`)
+## Post-release distribution
 
-The standing `uv-secure` gate in `maintain.sh` is clean; this is the changelog of how each alert was resolved, so a future alert is not re-triaged from scratch.
+`.github/workflows/distribute.yml` runs on the same published-release event. It
+waits for the matching source distribution to appear on PyPI, then:
 
-- **idna** bumped 3.11 -> 3.16, fixing GHSA-65pc-fj4g-8rjx.
-- **aiohttp** bumped 3.13.5 -> 3.14.0 via `uv lock --upgrade-package aiohttp`, fixing GHSA-hg6j-4rv6-33pg + GHSA-jg22-mg44-37j8.
-- **basicsr** Dependabot alert GHSA-86w8-vhw6-q9qq is resolved by removal: the experimental `restore` extra was retired and basicsr is no longer anywhere in the dependency tree.
-- **torch** Dependabot alert **GHSA-rrmf-rvhw-rf47** (`torch.jit.script` memory corruption, alert range `<= 2.12.1`) was dismissed `not_used` on 2026-06-10 (torch is a transitive dep of the optional `gpu` extra only and the codebase never calls `torch.jit`) and **resolved by upgrade on 2026-07-21**: the lock carries torch **2.13.0**, above the patched floor, so `uv-secure` is clean. If the GitHub alert has not auto-closed on the lock bump, close it manually as fixed.
-- **setuptools** bumped 81.0.0 -> 83.0.0 (2026-07-21), fixing PYSEC-2026-3447.
+- updates the Homebrew tap formula URL and SHA-256;
+- triggers a factory rebuild of the Hugging Face Space.
+
+The workflow can also be started manually with an optional version input.
+Conda-forge updates are outside this workflow.
+
+If a distribution job fails because a repository or Hugging Face credential is
+invalid, rotate the corresponding GitHub secret and rerun the failed job. A
+manual Homebrew formula update is the fallback when its automation is blocked.
+
+## Source distribution boundary
+
+The wheel includes the package under `src/`.
+
+The source distribution explicitly excludes `/data` through
+`[tool.hatch.build.targets.sdist]` in `pyproject.toml`. Keep that exclusion:
+calibration captures and test corpora do not belong in the published package
+archive.
+
+## Build backend
+
+The package uses hatchling through the unpinned `hatchling` build requirement in
+`pyproject.toml`. Uploading uses uv rather than the older twine-based action.
+
+## Other channels
+
+The repository includes a conda recipe under `packaging/conda/recipe.yaml`.
+Keep its runtime dependencies aligned with `pyproject.toml`.
+
+The ComfyUI nodes are maintained and versioned separately from this package.
+A library release does not by itself publish a new ComfyUI node version.
+
+## Release verification
+
+After publication, verify:
+
+- both wheel and source distribution exist on PyPI;
+- the package version matches the tag;
+- the Homebrew formula points to the new source distribution;
+- the distribution workflow completed successfully;
+- a clean install can run `remove-ai-watermarks --version`.

docs/research-doubao-distillation.md

@@ -1,20 +1,32 @@
 # Doubao clean-reverse-alpha distillation (re-investigated 2026-05-29)
 
+> Research archive. Reverse-alpha pixel recovery is no longer part of the
+> current visible-removal pipeline. The current implementation uses
+> localize-then-fill; see `docs/module-internals.md`.
+
 > Relocated verbatim from `CLAUDE.md` on 2026-06-11 to keep the always-loaded
 > context small. Long single-line entries were reformatted into paragraphs;
 > no content was changed or summarized.
 
-**RESOLVED 2026-05-29: black+gray Doubao captures were obtained and a reverse-alpha is built** (`doubao_engine.remove_watermark_reverse_alpha`, `assets/doubao_alpha.png`; see the `doubao_engine.py` section in `docs/module-internals.md`). The captures (`data/doubao_capture/captures/`, now committed) confirmed the alpha-composite model: on black `captured = a*logo`, logo pure white.
+**RESOLVED 2026-05-29: black+gray Doubao captures were obtained and a reverse-alpha was built.**
+That historical method, `doubao_engine.remove_watermark_reverse_alpha`, has since
+been removed. Its detection silhouette remains at
+`src/remove_ai_watermarks/assets/doubao_alpha.png`. The committed captures in
+`data/calibration/doubao/` confirmed the alpha-composite model: on black
+`captured = a*logo`, logo pure white.
 
 **UPDATE 2026-05-31 (issue #13 follow-up): the first build was NOT "exact"** — it left a readable "豆包AI生成" outline on the real sample (the detector was fooled, conf 0.0). The alpha is now rebuilt by `scripts/visible_alpha_solve.py` (the careful gray-self solve shared with Jimeng), removal always-aligns + thin-inpaints, and the locate box was widened; see the `doubao_engine.py` section in `docs/module-internals.md`. The notes below (the failed content-image distillation) are retained as the record of why controlled captures were necessary.
 
 **Conclusion (historical): pure reverse-alpha distilled from content images does NOT work, and the blocker is the WRONG kind of data, not too little of it.**
 
-The earlier framing ("need ~5-8 PRISTINE same-resolution originals") is obsolete -- a local corpus of pristine originals holds plenty. Curate them with `DoubaoEngine.detect` + an NCC filter against a clean glyph template, keeping only marks at offset ≈ (0,0): that yields e.g. **15 pixel-aligned 2048² marks** (sub-pixel drift, not the ±50 px the old lossy/mixed-res scrapes had), plus 1086x1448 / 1792x2400 clusters. With those, LaMa-clean `O` + weighted-LS (and per-pixel I-on-O regression) for `α` (+ logo color) was tried end-to-end and **still leaves a persistent ghost outline.**
+Curate same-resolution originals with `DoubaoEngine.detect` and an NCC filter against
+a clean glyph template, keeping only aligned marks. Even with aligned inputs,
+LaMa-clean `O` plus weighted least squares and per-pixel regression for `α` and logo
+color still leaves a persistent ghost outline.
 
 Diagnosed why, empirically (cached stacks, `/tmp/doubao_distill`): (1) the mark is a clean white overlay with **no dark halo** -- over glyph pixels ~54% are brighter than the clean bg, only ~4% darker -- so the white-logo model `I=(1-α)O+α·255` is correct; (2) but content backgrounds are almost never dark *under* the mark (median darkest available bg over glyph pixels = **58/255**; only ~13% of mark pixels are ever observed on a bg < 40), so on bright backgrounds the equation is ill-conditioned and `α` is unidentifiable; (3) LaMa's `O` is a plausible **hallucination**, not the true pre-mark background, which compounds the error, and per-pixel regression on ~15 obs overfits into color noise.
 
-**Why Gemini's engine is clean (verified in GeminiWatermarkTool `src/core/watermark_engine.cpp`): its alpha map is the watermark stamped on a PURE-BLACK background**, where `watermarked = α·255 + (1-α)·0 = α·255`, so `alpha = capture/255` exactly -- no estimation. (`gemini_bg_*.png` is literally the sparkle in grey on black.) So the real Doubao unlock is the same controlled capture, **not more content images**. Black/white/gray seeds exist (`data/doubao_capture/seeds/seed_*_1x1_2048x2048.png`); a capture run (feed a black seed through doubao.com edit mode, download the *original*) was requested from the #13 reporter 2026-05-29. With ~2-3 black captures we get `α = capture/255` for free, Gemini-quality.
+**Why Gemini's engine is clean (verified in GeminiWatermarkTool `src/core/watermark_engine.cpp`): its alpha map is the watermark stamped on a PURE-BLACK background**, where `watermarked = α·255 + (1-α)·0 = α·255`, so `alpha = capture/255` exactly -- no estimation. (`gemini_bg_*.png` is literally the sparkle in gray on black.) So the real Doubao unlock is the same controlled capture, **not more content images**. The retained black and gray outputs live in `data/calibration/doubao/`; local solid-color seeds are regenerable and are not committed.
 
 **Until black captures arrive, the shipped direction is precise canonical glyph mask + inpaint (cv2 default, lama optional), NOT reverse-alpha.**
 

docs/supported-signals.md

@@ -0,0 +1,127 @@
+# Supported signals
+
+This page describes the current support boundary. A check mark means that the
+repository contains a corresponding code path. It does not guarantee detection
+or removal on every future vendor version.
+
+## Visible marks
+
+The `visible` command registers these mark keys:
+
+| Key | Mark | Expected area | Important limit |
+| --- | --- | --- | --- |
+| `gemini` | Google Gemini sparkle | Usually bottom right | Detection includes a false positive gate. |
+| `doubao` | `豆包AI生成` | Bottom right | Vendor specific text detector. |
+| `jimeng` | `★ 即梦AI` | Bottom right | Vendor specific text detector. |
+| `qwen` | `千问AI生成` | Bottom right | Strict visual gate. |
+| `kling` | `可灵AI 3.0` | Bottom right | Only calibrated variants are covered. |
+| `yuanbao` | `元宝` over `AI生成` | Bottom right | Standard two-line variant only. |
+| `samsung` | `✦ Contenuti generati dall'AI` | Bottom left | Calibrated for the Italian text variant. |
+| `runninghub` | `RunningHub AI生成` | Top left | Strict visual and position gates. |
+| `baidu` | `百度 AI生成` | Bottom right | Detector and extended removal footprint. |
+| `liblib` | `LibLibAI` | Bottom center | Includes a minimum image size gate. |
+| `jimeng_pill` | `AI生成` pill | Top left | Weak detector with additional product and background gates. |
+
+`--mark auto` evaluates all registered marks and removes every selected match.
+Known marks are localized to a mask, then the selected fill backend reconstructs
+the masked area.
+
+Marks from other vendors are not detected automatically. Use `erase --region`
+when you can select the affected area yourself.
+
+## Fill backends
+
+| Backend | Install | Behavior |
+| --- | --- | --- |
+| `cv2` | Core package | Classical OpenCV inpainting |
+| `migan` | `remove-ai-watermarks[migan]` | MI-GAN through ONNX Runtime |
+| `lama` | `remove-ai-watermarks[lama]` | big-LaMa through ONNX Runtime |
+| `auto` | Depends on installed extras | Selects LaMa, then MI-GAN, then OpenCV |
+
+The learned backends download model files on first use.
+
+## Metadata and provenance
+
+The inspection and stripping code handles signals in these groups:
+
+- C2PA Content Credentials and supported cloud manifest references;
+- EXIF and XMP generator fields;
+- IPTC AI disclosure fields;
+- PNG text chunks and embedded generation parameters;
+- China TC260 AIGC labels in supported metadata placements;
+- xAI and Grok EXIF signature fields;
+- Samsung AI editing markers;
+- Hugging Face job metadata;
+- open Stable Diffusion style DWT-DCT watermarks with the `detect` extra;
+- Adobe TrustMark with the `trustmark` extra.
+
+`identify` combines detected signals into a `ProvenanceReport`. It reports
+unknown when evidence is absent. It never treats missing metadata as proof that
+an image is human made.
+
+## File and container formats
+
+Pixel based image commands discover these extensions:
+
+- PNG;
+- JPEG;
+- WebP;
+- HEIC and HEIF;
+- AVIF.
+
+Metadata inspection and removal additionally have container paths for:
+
+- JPEG XL metadata;
+- MP4, MOV, M4V, and M4A;
+- WebM, MKV, MKA, MP3, WAV, FLAC, OGG, OGA, Opus, and AAC when ffmpeg is
+  available.
+
+JPEG image metadata stripping removes targeted metadata segments without
+re-encoding the entropy coded image scan. PNG and WebP removal preserves pixel
+values through lossless output paths. HEIC, HEIF, AVIF, and other containers
+use their format specific paths.
+
+## Invisible watermarks
+
+The `invisible` command uses diffusion regeneration. It targets watermark
+patterns by changing the image rather than decoding and deleting a known
+payload.
+
+Current pipeline values:
+
+- `controlnet`;
+- `sdxl`;
+- `qwen`;
+- `qwen-zimage`;
+- legacy alias `default`, which resolves to `sdxl`.
+
+SynthID does not have a public local pixel decoder in this project. The tool can
+infer likely presence from supported provenance metadata, but after that
+metadata is removed a local negative result is inconclusive.
+
+The optional `detect` extra is different: it provides a local decoder for the
+open DWT-DCT watermark used by some Stable Diffusion, SDXL, and FLUX workflows.
+That signal is carrier and transformation sensitive, so a negative is still
+not a universal clean verdict.
+
+## Provider overview
+
+| Provider or family | Visible | Invisible path | Metadata or provenance |
+| --- | --- | --- | --- |
+| Google Gemini | Sparkle | Diffusion regeneration for SynthID | C2PA and related source signals |
+| OpenAI image generators | None registered | Diffusion regeneration for supported invisible signals | C2PA and generator provenance |
+| Stable Diffusion and SDXL | None registered | Diffusion regeneration; optional open decoder | Embedded parameters and text metadata |
+| FLUX | None registered | Diffusion regeneration; optional open decoder | C2PA for supported sources |
+| Adobe Firefly | None registered | No proprietary local decoder | C2PA; optional TrustMark decoder |
+| Midjourney | None registered | No registered pixel decoder | EXIF, XMP, and IPTC signals |
+| ByteDance generators | Doubao and Jimeng marks | No registered pixel decoder | TC260 AIGC and supported C2PA signals |
+| Qwen | Qwen mark | No registered pixel decoder | TC260 AIGC |
+| Kling | Kling mark | No registered pixel decoder | TC260 AIGC |
+| Baidu | Baidu mark | No registered pixel decoder | TC260 AIGC |
+| LibLibAI | LibLibAI mark | No registered pixel decoder | TC260 AIGC |
+| RunningHub | RunningHub mark | No registered pixel decoder | TC260 AIGC |
+| Samsung Galaxy AI | One locale specific mark | No registered pixel decoder | C2PA and Samsung markers |
+
+For detector thresholds, measured limits, and incident history, see
+[module internals](module-internals.md) and
+[known limitations](known-limitations.md).

docs/synthid-robust-identity-research-2026-06-08.md

@@ -1,5 +1,8 @@
 # Deep research: SynthID-safe face-identity recovery for SDXL (2026-06-08)
 
+> Research archive. This dated follow-up records evidence available during the
+> study. It is not the current command reference.
+
 **Stats:** {"angles": 6, "sourcesFetched": 28, "claimsExtracted": 104, "claimsVerified": 25, "confirmed": 19, "killed": 6, "afterSynthesis": 6, "urlDupes": 1, "budgetDropped": 7, "agentCalls": 111}
 
 ## Summary
@@ -49,7 +52,7 @@ Arc2Face README verbatim: 'Arc2Face is built upon SD1.5' with stable-diffusion-v
 **Vote:** 3-0 on the narrow factual claims (SDXL base + CLIP-G encoder)
 
 
-GitHub README explicitly instructs 'Download the pretrained base models from SDXL-base-1.0 and CLIP-G' (CLIP-ViT-bigG-14-laion2B-39B-b160k). Neither README nor arXiv 2406.07209 mention ArcFace/InsightFace/antelopev2/buffalo_l. Architecturally descended from IP-Adapter (CLIP-image-embedding family), not from FaceID/InstantID/PhotoMaker-V2. Verifier caveat (high confidence on the license-narrow claim, medium on suitability): CLIP-image face-ID accuracy ~80.95% vs specialized face recognition ~87.61% — license-safe but probably not identity-grade for portraits. Confidence is medium because the suitability claim for raiw.cc face-identity use case has not been validated empirically.
+GitHub README explicitly instructs 'Download the pretrained base models from SDXL-base-1.0 and CLIP-G' (CLIP-ViT-bigG-14-laion2B-39B-b160k). Neither README nor arXiv 2406.07209 mention ArcFace/InsightFace/antelopev2/buffalo_l. Architecturally descended from IP-Adapter (CLIP-image-embedding family), not from FaceID/InstantID/PhotoMaker-V2. Verifier caveat (high confidence on the license-narrow claim, medium on suitability): CLIP-image face-ID accuracy ~80.95% vs specialized face recognition ~87.61% — license-safe but probably not identity-grade for portraits. Confidence is medium because suitability for portrait and group-photo inputs has not been validated empirically.
 
 - https://proceedings.iclr.cc/paper_files/paper/2025/file/ed4df1609bf7d8602435341c9ce2ab5f-Paper-Conference.pdf
 - https://github.com/MS-Diffusion/MS-Diffusion
@@ -83,7 +86,7 @@ Six claims were refuted in adversarial verification, two of them load-bearing: A
 
 ## Open questions
 
-- Does MS-Diffusion (or any CLIP-image-embedding SDXL adapter) achieve usable face-identity fidelity on the raiw.cc input distribution (portraits + group photos), or is the ArcFace gap (~7 pp face-ID accuracy) visually disqualifying — and can a face-specific CLIP fine-tune close it?
+- Does MS-Diffusion (or any CLIP-image-embedding SDXL adapter) achieve usable face-identity fidelity on portraits and group photos, or is the ArcFace gap (~7 pp face-ID accuracy) visually disqualifying — and can a face-specific CLIP fine-tune close it?
 - Has InstantX (or any community fork) actually shipped an InstantID variant retrained on a commercially-licensed face embedder since the maintainer's 2024 commitment, and if so what is its identity-fidelity vs the antelopev2 original?
 - What is the exact diffusers-0.38 compat status of InstantID, MS-Diffusion, and PuLID-FLUX inference scripts — does any need a fork the way PhotoMaker-V1 did, and if so what specifically breaks?
 - Is there a single-pipeline multi-subject identity-preservation method (mask-guided regional ID-adapters, multi-subject InstantID, MS-Diffusion multi-subject mode) that handles group photos without the per-face crop+composite patchwork that PhotoMaker-V2 produced?
@@ -161,7 +164,7 @@ on Modal A100 in two phases:
    `ip_adapter_scale=1.0`, `controlnet_scale=1.0` brought identity closer to
    original but introduced more "SDXL gloss / clean skin" aesthetic.
 
-**Net finding for raiw.cc (load-bearing).** The fundamental issue is structural:
+**Net finding for a commercial deployment (load-bearing).** The fundamental issue is structural:
 ArcFace encodes "this person's general look" (ethnicity, gender, basic facial
 geometry) at 512 dimensions; SDXL decodes that embedding into pixels with the
 inherent SDXL aesthetic (smooth skin, symmetric pores, AI-photoreal look).
@@ -182,4 +185,4 @@ for embedding-driven regeneration and makes the face read as "AI-generated"
 rather than "the original person". The `instantid` and `photomaker` extras
 stay in the library as opt-in for research / personal use where users
 explicitly want identity regeneration; the CLI flag and module docstrings
-state the trade-off at every entry point.
+state the trade-off at every entry point.

docs/synthid-robust-identity-research.md

@@ -1,5 +1,10 @@
 # SynthID-robust face identity for an SDXL removal pipeline (research)
 
+> Research archive. This document evaluates approaches considered during the
+> study. It does not describe the current public interface. The current
+> pipelines and limits are documented in `docs/cli.md` and
+> `docs/known-limitations.md`.
+
 > **Status (2026-06-08): retired.** Every approach described below was empirically
 > tested and rejected -- see `docs/synthid-robust-identity-research-2026-06-08.md`
 > "Empirical follow-up" for the final conclusion. The library no longer ships any
@@ -12,7 +17,7 @@ canny-ControlNet watermark-removal pipeline (denoise 0.20-0.30) is BOTH (a)
 commercial-safe end-to-end and (b) does not re-introduce the SynthID pixel
 watermark the removal pass just destroyed?
 
-**Constraint.** raiw.cc is a paid service, so every component (adapter weights AND
+**Constraint.** A paid deployment requires every component (adapter weights AND
 the face embedder it conditions on AND any base model) must be Apache-2.0 / MIT /
 BSD or otherwise clearly commercial-permitted. Non-commercial is disqualifying.
 
@@ -61,7 +66,7 @@ extra). V2 uses a DUAL ID encoder (CLIP image features + ArcFace embedding),
 which delivers true identity-from-embedding face regeneration. The cost is that
 the ArcFace embedding comes from InsightFace's `antelopev2`/`buffalo_l` model
 packs, which are released under a non-commercial / research-only license. **So
-the shipped restore path is NON-COMMERCIAL.** raiw.cc and any other monetized
+the shipped restore path is NON-COMMERCIAL.** Any monetized
 deployment must NOT install the `photomaker` extra. The CLI flag and module
 docstring both call this out at every entry point.
 
@@ -213,9 +218,9 @@ from the test set + this doc).
   - canny ControlNet conditioning can fight the ID embedding (edges of the
     ORIGINAL face vs identity of the SAME person regenerated) -- expect to tune
     `controlnet_conditioning_scale` down a notch on photoreal faces;
-  - PhotoMaker was trained on a celebrity-skew distribution; real-user faces
+  - PhotoMaker was trained on a celebrity-skew distribution; representative faces
     (especially non-white, non-Western, elderly, children) may have lower
-    fidelity. Measure on the real upload distribution.
+    fidelity. Measure on a representative evaluation distribution.
 
 ## 6. Integration cost (rough)
 
@@ -253,7 +258,7 @@ from the test set + this doc).
    per-vendor oracles. Expected: SynthID cleared (the regeneration is the same)
    AND identity recovered (the embedding adds it back).
 4. **Honest exit criteria.** Ship only if BOTH oracle reads clean AND a small
-   user-perception test on real uploads says "looks like me". If identity is
+   user-perception test on representative inputs says "looks like me". If identity is
    still too soft on small faces -> add stacked-reference (multiple crops of the
    same upload at different scales) before reaching for a non-commercial
    embedder.

docs/synthid.md

@@ -1,5 +1,10 @@
 # SynthID-Image: technical reference
 
+> Technical research reference. Current package behavior is defined by the
+> [supported signals](supported-signals.md), [known limitations](known-limitations.md),
+> and [module internals](module-internals.md). Dated measurements below are
+> historical evidence and should not be read as current CLI defaults.
+
 This document covers how Google SynthID for images works mechanically, what it
 survives, what removes it, and the current deployment landscape. It is written
 for engineers working on watermark detection and removal -- specifically to
@@ -175,11 +180,11 @@ A controlled study (June 2026, clean v0.8.6 with text/face protection OFF,
 native resolution on this repo's default SDXL pipeline) measured the minimum
 img2img strength that removes the SynthID pixel watermark, verified per image on
 the vendor's own oracle (openai.com/verify for OpenAI, the Gemini app "Verify
-with SynthID" for Google). Each subject is archived in `data/synthid_corpus/` as a
-pos original plus its minimum-clearing cleaned output (manifest `verified_via` =
-`openai-verify` / `gemini-app`), EXCEPT one third-party image from issue #14, which
-was oracle-verified but is not committed (third-party content stays out of the
-public corpus).
+with SynthID" for Google). The reusable originals are stored once in
+`data/synthid/originals/`, with their input verification in `manifest.csv`.
+Generated cleaned outputs are not committed; the table below is the durable
+record of the historical oracle verdicts. One third-party image from issue #14
+was oracle-verified but is not committed.
 
 **Oracle validation order: start with OpenAI.** When validating removal across
 vendors, run the OpenAI arm first. `openai.com/verify` is more accessible than the
@@ -431,27 +436,14 @@ conditioning, never by copying original pixels.**
   OpenAI and Gemini oracles. This is a quality recommendation for the measured content,
   not broad removal certification; very small text can still degrade.
   See `docs/known-limitations.md` for the metrics, runtime, and validation scope.
-- **Face identity:** canny holds face *structure* but not *identity*. Shipped as the
-  optional `--restore-faces` GFPGAN post-pass (`face_restore.py`, the `restore`
-  extra, experimental/opt-in, off by default). It runs GFPGAN on the ORIGINAL
-  faces and feather-composites the restored face REGIONS into the cleaned image.
-  **WARNING (oracle-confirmed 2026-06-04): this pass can RE-INTRODUCE SynthID into
-  the face regions -- the earlier "GFPGAN re-synthesizes from a StyleGAN2 prior ->
-  scrubs SynthID -> oracle-confirmed clean" claim was WRONG.** At the default fidelity
-  weight `0.5` GFPGAN blends ~half the ORIGINAL (watermarked) face pixels with the
-  prior, and SynthID is robust to that partial blend, so the composited face carries
-  the watermark back in -- over the diffusion-cleaned face. Confirmed by a clean A/B:
-  `gemini_3` read SynthID-detected after controlnet @ 0.20/0.25 WITH restore, but
-  NOT-detected after the same controlnet @ 0.20 with `--no-restore-faces` (only
-  restore differed). Content-dependent (a second face image cleared WITH restore),
-  which is why a single-image check earlier read "clean". **Fix directions (not yet
-  done): run GFPGAN on the diffusion-CLEANED image not the original; or drop the
-  weight well below 0.5; or leave restore OFF for removal -- each needs oracle
-  re-validation.** Commercial-
-  safe (GFPGAN Apache-2.0 + RetinaFace MIT); the CodeFormer alternative is
-  NON-COMMERCIAL and is not shipped. (An IP-Adapter FaceID approach was tried and
-  REMOVED -- it needs high denoise strength and corrupts faces at removal strength;
-  see `docs/controlnet-removal-pipeline-research.md`.)
+- **Face identity:** canny holds face *structure* but not *identity*. The standard
+  SDXL and ControlNet profiles do not run a separate face-restoration option.
+  Earlier GFPGAN, PhotoMaker, and FaceID experiments were removed after they
+  degraded identity or risked reintroducing source pixels. The separate
+  `qwen-zimage` profile now provides the only shipped face-specific stage:
+  YuNet and SAM locate faces, then Z-Image regenerates the selected original
+  face crops before a feathered composite. See
+  `docs/controlnet-removal-pipeline-research.md` for the historical experiments.
 
 ### 5.2 Strength setting
 
@@ -536,7 +528,7 @@ manifest get `OPENAI_STRENGTH` 0.10, the one without C2PA falls to
 | image | content type | size | strength | `--auto`/controlnet | `default` |
 |---|---|---|---|---|---|
 | typography card | flat text | 1122x1402 | 0.10 | clean | clean |
-| raiw.cc poster | flat graphic (logo + flat fills) | 1024x1536 | 0.10 | clean | **detected** |
+| Flat poster | flat graphic (logo + flat fills) | 1024x1536 | 0.10 | clean | **detected** |
 | 9-face grid | photoreal | 1448x1086 | 0.10 | **detected** | clean |
 | bracelet product photo | photoreal | 1600x1600 | 0.15 | **detected** | clean |
 
@@ -575,16 +567,18 @@ content×pipeline table above conflates a borderline/non-deterministic 0.15 resu
 with deterministic content behavior -- the photoreal-survives-controlnet effect is
 solid at 0.10 but at 0.15 it is near-threshold noise; (2) for reliable removal pick
 a strength with MARGIN above the borderline (controlnet >= 0.20), not exactly on
-it; (3) **engineering follow-up for raiw.cc: the controlnet pipeline should use a
-HIGHER vendor strength than `default` (it currently shares `resolve_strength`) --
-e.g. controlnet floor 0.20 -- calibrated per vendor/content on the GPU worker where
-batches are cheap. The shared 0.10/0.15 is tuned for `default`, not controlnet.**
+it; (3) **historical engineering conclusion:** this dated run argued for a
+higher ControlNet strength than the then-current default. That proposal was
+later superseded. The current resolver intentionally shares the 0.10/0.15
+ladder between SDXL and ControlNet and uses a separate Qwen ladder; see
+`noai/watermark_profiles.py`.
 Source images are private (faces / product shots), not committed; reproduce on any
 photoreal + flat-graphic gpt-image pair, varying the seed, and re-checking the
 oracle.
 
 **Gemini pass + the face-restore re-introduction (2026-06-04).** Four Gemini
-originals via `--auto` (controlnet) at `--max-resolution 1024`, checked on the
+originals via the then-current `--auto` ControlNet path at `--max-resolution 1024`,
+checked on the
 Gemini "Verify with SynthID" oracle (Google content needs the Google oracle, not
 openai.com/verify):
 - Most cleared at controlnet 0.15-0.25; `gemini_3` (a large central FACE, +restore)
@@ -599,23 +593,21 @@ openai.com/verify):
   robust to downscaling by design, and the study's resolution trend says LOWER
   processing res needs LESS strength, so 1024 was never the wall.)
 
-**Certified controlnet floors (isolated Modal GPU sweep + oracle,
+**Historical controlnet certification, superseded by the current vendor-adaptive
+defaults (isolated GPU sweep + oracle,
 restore OFF, <= 1536, each vendor on its own oracle):** OpenAI **0.20** (2 photoreal x
 seed {1,2,3} = 6/6 clean; the 0.15-flipper is seed-robust at 0.20) and Gemini **0.30**
 (0.20 detected -> 0.30 clean on 2/2 seeds). OpenAI 0.20 transfers to prod
 (resolution-independent); Gemini 0.30 holds only <= 1536 -- Gemini is
-resolution-sensitive and raiw.cc runs NATIVE, so cap Gemini <= 1536 + use 0.30 or
+resolution-sensitive, so a native-resolution deployment should cap Gemini <= 1536 + use 0.30 or
 native-calibrate (~0.35+). See `docs/controlnet-removal-pipeline-research.md` for the
 table.
 
-**Net for raiw.cc:** (1) controlnet needs a higher, per-vendor strength than
-`default` -- CERTIFIED OpenAI 0.20 / Gemini 0.30 (above); add a controlnet-specific
-schedule to `resolve_strength`, do not reuse the default ladder; (2) the
-`--restore-faces` pass is now SynthID-safe by construction (the GFPGAN-on-original
-path that re-added SynthID was removed 2026-06-04; the shipped restore is
-PhotoMaker-V2, NON-COMMERCIAL, see `photomaker_restore.py`); (3)
-removal near threshold is seed-non-deterministic -> FIX the prod seed (kills the
-coin-flip; ship a deterministic certified config).
+**Current implication:** the old floor table remains evidence about the dated
+test set, not the current resolver. The shipped SDXL and ControlNet defaults are
+defined in `watermark_profiles.py`, and face restoration is available only
+through the separate `qwen-zimage` profile. Removal near a threshold remains
+seed dependent, so reproducible verification requires a fixed seed.
 
 ---
 

docs/text-protection-research.md

@@ -1,5 +1,8 @@
 # Text protection research: crisp text under a "watermark removed everywhere" constraint
 
+> Research archive. This page records evaluated ideas, including rejected
+> prototypes. Current behavior is documented in the user guides and source.
+
 Date: 2026-05-29. Source: a deep-research run (104 agents, 5 search angles, sources
 fetched and 3-vote adversarially verified). Not committed automatically — saved as a
 research note for the next session.

docs/watermarking-landscape.md

@@ -1,47 +1,69 @@
 # Watermarking landscape (research 2026-05-24)
 
-> Relocated verbatim from `CLAUDE.md` on 2026-06-11 to keep the always-loaded
-> context small. Long single-line entries were reformatted into paragraphs;
-> no content was changed or summarized.
+> Research and signal inventory. Code-facing statements are updated with the
+> implementation; dated vendor observations remain snapshots and may change
+> independently of the package.
 
 Who embeds what, and whether it is locally detectable (so we know which gaps are fillable). See `identify.py` for what we read.
-- **Locally detectable (open decoder, no key/API):** Stable Diffusion / SDXL / FLUX via `imwatermark` DWT-DCT (now covered by `invisible_watermark.py`). FLUX uses the same library (`black-forest-labs/flux2` `src/flux2/watermark.py`, 48-bit `0b001010101111111010000111100111001111010100101110`); SDXL is the diffusers `WATERMARK_MESSAGE` (`0b101100111110110010010000011110111011000110011110`). **Caveat: the `imwatermark` dwtDct decode is carrier-fragile on a broad class of real images, NOT just re-encode-fragile, and it is a POSITIVE-ONLY signal.** A clean encode->decode round-trip (no re-encode at all) recovers 48/48 bits on some carriers (random noise, chatgpt-1.png 48/48, firefly-1.png 45/48) but FAILS on many others — verified 2026-06-19 that a *known-embedded* watermark only round-trips 28-39/48 (below the safe `_MATCH_48` = 44 gate, random baseline ~24) on the FLUX fox sample (28), doubao-1.png (39), a 1024² minimalist-flat FLUX image (28), AND a **clean synthetic bright-flat fill with NO watermark at all (28)**. The failure does NOT track texture (firefly lapvar ~11 passes; the flat FLUX lapvar ~56 fails); it correlates with a degenerate decode where the raw bits read **all-ones (48/48 ones)** — which a clean synthetic image reproduces, so **all-ones is a CARRIER ARTIFACT, NOT a watermark signal** (a double-embed test also showed a pre-existing embed does not corrupt a second embed — no interference). Net: trust a `detect_invisible_watermark` hit, but treat a `None`/no-match as **inconclusive** whenever a positive-control embed on the same carrier does not first recover >=44/48. The 44 gate is a deliberate precision choice (lowering it would admit false positives).
+- **Locally detectable (open decoder, no key/API):** Stable Diffusion / SDXL / FLUX via `imwatermark` DWT-DCT (now covered by `invisible_watermark.py`). FLUX uses the same library (upstream `black-forest-labs/flux2`, file `src/flux2/watermark.py`, 48-bit `0b001010101111111010000111100111001111010100101110`); SDXL is the diffusers `WATERMARK_MESSAGE` (`0b101100111110110010010000011110111011000110011110`). **Caveat: the `imwatermark` dwtDct decode is carrier-fragile on a broad class of real images, NOT just re-encode-fragile, and it is a POSITIVE-ONLY signal.** A clean encode->decode round-trip (no re-encode at all) recovers 48/48 bits on some carriers (random noise, chatgpt-1.png 48/48, firefly-1.png 45/48) but FAILS on many others — verified 2026-06-19 that a *known-embedded* watermark only round-trips 28-39/48 (below the safe `_MATCH_48` = 44 gate, random baseline ~24) on the FLUX fox sample (28), doubao-1.png (39), a 1024² minimalist-flat FLUX image (28), AND a **clean synthetic bright-flat fill with NO watermark at all (28)**. The failure does NOT track texture (firefly lapvar ~11 passes; the flat FLUX lapvar ~56 fails); it correlates with a degenerate decode where the raw bits read **all-ones (48/48 ones)** — which a clean synthetic image reproduces, so **all-ones is a CARRIER ARTIFACT, NOT a watermark signal** (a double-embed test also showed a pre-existing embed does not corrupt a second embed — no interference). Net: trust a `detect_invisible_watermark` hit, but treat a `None`/no-match as **inconclusive** whenever a positive-control embed on the same carrier does not first recover >=44/48. The 44 gate is a deliberate precision choice (lowering it would admit false positives).
 
   **Root cause and external confirmation (deep-research 2026-06-19, adversarially verified).** This is the SCHEME's ceiling, not our usage — there is no better decoder to adopt. The imwatermark maintainers state verbatim (both the ShieldMnt and Stability-AI READMEs) that the algorithm "cannot guarantee to decode the original watermarks 100% accurately even though we don't apply any attack." Independent measurement (WMAdapter, arXiv:2406.08337 Table 2) puts dwtDct at only **~0.79 bit accuracy on CLEAN images (~38/48 bits — already below our 44 gate)**, collapsing to ~0.50 (chance) under crop/JPEG. Two code-verified + locally-reproduced mechanisms drive the content-dependent failures: (1) the decoder reads each bit as the **highest-magnitude DCT coefficient per block**, so any content coefficient exceeding the encoded target flips the bit; (2) the default embed is in the **YUV chroma channel, which 8-bit-clamps on white/bright pixels** (a +36 chroma delta survives a white-fill round-trip as only +4, ~89% loss) — this is the mechanism behind the bright-flat / minimalist failures and the all-ones degenerate decode. No maintained fork or detector decodes this scheme reliably: the WAVES benchmark (arXiv:2401.08573) relegates DWT-DCT to supplementary appendix G.5 and targets Stable Signature / Tree-Ring / StegaStamp instead; learned encoder/decoder schemes reach ~0.98-0.99 clean but are a DIFFERENT watermark class (not what SDXL/FLUX stamp). `dwtDctSvd` does not help (SDXL embeds `dwtDct`; dwtDctSvd cannot decode it, and its clean accuracy ~0.72 is lower). **Authoritative conclusion: the open DWT-DCT mark cannot be turned from positive-only into a reliable real-world detector; keep it positive-only and rely on C2PA.** (Refuted along the way: that the library is unmaintained, and that it is robust to JPEG but only fails on geometric attacks — both did not survive verification.)
 
   Consequence for the FLUX hosted-output question (BFL Playground, FLUX.2 [pro] + FLUX.1 [dev], 2026-06-19): all samples carry the signed C2PA manifest (issuer "Black Forest Labs"); the open DWT-DCT decode returned `None`, but every available FLUX carrier (textured fox AND a minimalist-flat generation) failed the positive control (28/48), so the detector is blind on them and **whether BFL hosted output embeds the open pixel watermark is UNRESOLVED** (an earlier note here wrongly asserted it absent — overstated; a later note blamed "high texture" — also wrong, flat carriers fail too). What IS established: C2PA is the reliable FLUX identifier; the `_BITS_48` pattern is correct (round-trips on chatgpt/firefly/random). Resolving the hosted question needs a hosted FLUX carrier that first passes a >=44/48 positive control, which neither a textured nor a flat prompt produced — low priority (the open mark is only a stripped-metadata fallback).
-- **C2PA / IPTC (covered by the issuer/marker scan):** OpenAI, Google, Adobe Firefly, Microsoft (Designer + **Bing Image Creator** — collected 2026-05-24; Bing now runs Microsoft's own **MAI-Image** model, signs C2PA as "Microsoft", NOT OpenAI/DALL-E), **Stability AI** (collected from Brand Studio / DreamStudio successor; signs C2PA as "Stability AI Ltd", no SynthID, no imwatermark on its current Stable Image model — issuer added to `C2PA_ISSUERS`), and **Canva** (Magic Media signs C2PA as "Canva" + `trainedAlgorithmicMedia` with a generic `c2pa-rs` claim generator, no SynthID — issuer `b"Canva"` → "Canva (Magic Media)"; found on real production traffic 2026-06-19, which **disproved the earlier assumption** that Canva downloads are re-encoded exports that always strip C2PA). Still unsampled: Getty, Shutterstock. Midjourney embeds NO C2PA and no invisible watermark (our `mj-*` sample carried only the IPTC tag).
+- **C2PA / IPTC (covered by the issuer/marker scan):** OpenAI, Google, Adobe Firefly, Microsoft (Designer + **Bing Image Creator** — collected 2026-05-24; Bing now runs Microsoft's own **MAI-Image** model, signs C2PA as "Microsoft", NOT OpenAI/DALL-E), **Stability AI** (collected from Brand Studio / DreamStudio successor; signs C2PA as "Stability AI Ltd", no SynthID, no imwatermark on its current Stable Image model — issuer added to `C2PA_ISSUERS`), and **Canva** (Magic Media signs C2PA as "Canva" + `trainedAlgorithmicMedia` with a generic `c2pa-rs` claim generator, no SynthID — issuer `b"Canva"` → "Canva (Magic Media)"; verified samples disproved the earlier assumption that Canva downloads always strip C2PA). Still unsampled: Getty, Shutterstock. Midjourney embeds NO C2PA and no invisible watermark (our `mj-*` sample carried only the IPTC tag).
 
-**Samsung Galaxy AI** (Generative Edit / Sketch to Image / Portrait Studio on Galaxy S23 FE / S24 / S25, One UI 7+) signs C2PA as "Samsung Galaxy" with the standard `trainedAlgorithmicMedia` source type AND a proprietary `genAIType` marker; verified on real signed files 2026-05-29 (the standard scan catches the source type; `genAIType` additionally catches a Galaxy S24 file that omits it). It ALSO burns a **visible** localized wordmark into the pixels — a sparkle + "generated with AI" string in the bottom-LEFT corner (issue #37; the Italian "✦ Contenuti generati dall'AI" variant is calibrated) — removed by `samsung_engine.py` / `visible --mark samsung` (reverse-alpha, see the engine bullet); detection feeds `identify` as the medium `visible_samsung` signal. The string is locale-specific, so each locale needs its own captured alpha template.
+**Samsung Galaxy AI** signs supported edits with C2PA and may carry the
+proprietary `genAIType` marker. The registered visible detector covers the
+Italian `✦ Contenuti generati dall'AI` bottom-left variant. Removal follows the
+same localize-then-fill path as other registered text marks. Other locales and
+icon-only variants need separate calibrated silhouettes.
 
 **ASUS Gallery** also signs edited photos as C2PA (`com.asus.gallery`) but with no AI source type — a signer, not an AI marker.
 
 **Black Forest Labs (FLUX)** API output signs C2PA: `claim_generator_info "Black Forest Labs API"` + a `c2pa.ai_generated_content` assertion + `trainedAlgorithmicMedia` (issuer `b"Black Forest Labs"` added to `C2PA_ISSUERS`, platform "Black Forest Labs (FLUX)").
 
-**ByteDance Volcano Engine (Volcengine)** — the cloud behind Doubao / Jimeng — signs its AI image output with a cert from `certificate_center@volcengine.com` + `trainedAlgorithmicMedia` (issuer `b"volcengine"` → "ByteDance (Volcano Engine)", platform "ByteDance (Doubao / Jimeng / Volcano Engine)"); note this is the C2PA-signed surface, distinct from the XMP/PNG TC260 `AIGC` label Doubao also uses. All three verified on real signed files 2026-05-29. ByteDance's **international brand (BytePlus / Seedream / Seededit)** signs the SAME content as **"Byteplus Pte. Ltd."** — the bare `volcengine` needle missed it, so real BytePlus output was mis-attributed to "Adobe Firefly" (an incidental "Adobe XMP" toolkit string in the file's XMP, picked up by the fallback byte-scan once the clean manifest issuer matched nothing). Added issuer `b"Byteplus"` → org "BytePlus (ByteDance)" (platform resolves to the shared "ByteDance (Doubao / Jimeng / Volcano Engine)" label via the common `ByteDance` needle) so the clean manifest issuer attributes it directly; found on real production traffic 2026-06-19. ByteDance's consumer app **Dreamina** (the international Jimeng brand) signs as **"Bytedance Pte. Ltd."** with a `Dreamina/x.y` claim generator but, unlike the Volcano Engine surface, ships **NO `trainedAlgorithmicMedia`** — the generator name is the only AI signal, and the active manifest is frequently a plain `c2pa-tool` transcode with the real `Dreamina` token on an ingredient manifest. Added issuer `b"Dreamina"` → org "ByteDance (Dreamina)" with **`asserts_ai=True`** (see `constants.py`): the caBX / store-JSON byte-scan sees the token across all manifests, and the identity-AI flag lifts the AI verdict without a source-type. Mined from the retained corpus 2026-07 (7 files read `unknown` before, all now ByteDance). Registering the **issuer** `b"Bytedance Pte"` was deliberately AVOIDED — that same Singapore entity also signs non-AI CapCut edits (`CapCut/x.y` generator, `c2pa.created`, no AI marker), which must stay unattributed per the editor-vs-generator line; keying on the `Dreamina` generator token is precise.
-- **EXIF/XMP/PNG-text generator tag (caught by `exif_generator`):** **Ideogram** writes EXIF `Make="Ideogram AI"` (collected 2026-05-24 — no C2PA, no SynthID, no imwatermark; the Make tag is the only signal). Three more mined from the retained corpus 2026-06-22, all no-C2PA generator stamps that previously read as no-signal: **NovelAI** (anime SD) writes its stamp in PNG `tEXt` chunks `Software="NovelAI"` / `Source="NovelAI Diffusion V4.5 <hash>"` / `Title="NovelAI generated image"` — so `exif_generator` now reads PNG text chunks (`Software`/`Source`/`Title`/`Description`), not just EXIF/XMP; **Reve** (reve.com) writes EXIF `Software` / XMP `CreatorTool` = `reve.com` (token is the full `reve.com`, not bare `reve`, to avoid false-firing on "forever"/"reverie"); **Aphrodite AI** writes EXIF `Make`/`Software` = `Aphrodite AI`.
+**ByteDance Volcano Engine (Volcengine)** — the cloud behind Doubao / Jimeng — signs its AI image output with a cert from `certificate_center@volcengine.com` + `trainedAlgorithmicMedia` (issuer `b"volcengine"` → "ByteDance (Volcano Engine)", platform "ByteDance (Doubao / Jimeng / Volcano Engine)"); note this is the C2PA-signed surface, distinct from the XMP/PNG TC260 `AIGC` label Doubao also uses. ByteDance's **international brand (BytePlus / Seedream / Seededit)** signs the same content as **"Byteplus Pte. Ltd."**. The bare `volcengine` needle missed it, so BytePlus output was mis-attributed to "Adobe Firefly" through an incidental "Adobe XMP" toolkit string. Issuer `b"Byteplus"` now maps directly to "BytePlus (ByteDance)". ByteDance's consumer app **Dreamina** (the international Jimeng brand) signs as **"Bytedance Pte. Ltd."** with a `Dreamina/x.y` claim generator but, unlike the Volcano Engine surface, ships **no `trainedAlgorithmicMedia`**. Issuer `b"Dreamina"` maps to "ByteDance (Dreamina)" with **`asserts_ai=True`**. Registering the broader **issuer** `b"Bytedance Pte"` was deliberately avoided because that same entity also signs non-AI CapCut edits; keying on the `Dreamina` generator token is precise.
+- **EXIF/XMP/PNG-text generator tag (caught by `exif_generator`):** **Ideogram** writes EXIF `Make="Ideogram AI"` (collected 2026-05-24 — no C2PA, no SynthID, no imwatermark; the Make tag is the only signal). Additional verified generator stamps include **NovelAI** (`Software`, `Source`, and `Title` PNG text chunks), **Reve** (`Software` or XMP `CreatorTool` = `reve.com`), and **Aphrodite AI** (`Make` or `Software` = `Aphrodite AI`).
 - **xAI / Grok — its own EXIF signature scheme, NOT C2PA (DETECTED by `metadata.xai_signature`, built 2026-05-26).**
 
 Grok JPEG downloads (Aurora model) carry **no C2PA, no XMP, no SynthID, no IPTC** — only EXIF `Artist` = a UUID and EXIF `ImageDescription` = `Signature: <base64>` (a crypto signature, unverifiable locally without xAI's public key). This empirically kills the earlier unverified "xAI signs C2PA as xAI" lead — xAI is not even a C2PA member. `exif_generator` misses it (neither field holds an `AI_GENERATOR_TOKENS` token), so a dedicated detector `xai_signature(path)` matches the pair (`ImageDescription ~ ^Signature: [A-Za-z0-9+/=]{64,}` AND UUID `Artist`); wired into `has_ai_metadata`, `get_ai_metadata` (key `xai_signature`), and `identify` (signal `xai_signature`, platform "xAI (Grok / Aurora)").
 
-**Format confirmed stable across n=3 genuine generations:** exactly three EXIF tags (`Artist`, `ExifOffset`, `ImageDescription`), `Signature:` prefix constant, base64 payload 300-1004 chars. Two capture facts: (a) the `Artist` UUID **equals the public image id** in the asset URL (`https://imagine-public.x.ai/imagine-public/images/<uuid>.jpg`), so it is NOT a private per-user secret — only the `Signature` blob is; (b) the Grok web-UI image is a re-encoded **WebP with no signature** — the EXIF survives only in the *original* JPEG (download button or that public tokenless URL), which is why screenshots / re-encodes are metadata-stripped. A real fixture `data/samples/grok-1.jpg` plus **synthetic** JPEG fixtures (fake UUID + fake `Signature:` blob) cover the detector; never add a real Grok image carrying private content (the repo is public).
+**Format confirmed stable across n=3 genuine generations:** exactly three EXIF tags (`Artist`, `ExifOffset`, `ImageDescription`), `Signature:` prefix constant, base64 payload 300-1004 chars. Two capture facts: (a) the `Artist` UUID **equals the public image id** in the asset URL (`https://imagine-public.x.ai/imagine-public/images/<uuid>.jpg`), so it is NOT a private per-user secret — only the `Signature` blob is; (b) the Grok web-UI image is a re-encoded **WebP with no signature** — the EXIF survives only in the *original* JPEG (download button or that public tokenless URL), which is why screenshots / re-encodes are metadata-stripped. A real fixture `data/fixtures/provenance/grok-1.jpg` plus **synthetic** JPEG fixtures (fake UUID + fake `Signature:` blob) cover the detector; never add a real Grok image carrying private content (the repo is public).
 
-**Stripped on removal too:** `remove_ai_metadata` now calls `_scrub_ai_exif` on the JPEG EXIF, which deletes the xAI Signature+UUID-Artist pair **and** any `Software`/`Make`/`Artist`/`ImageDescription` tag holding an `AI_GENERATOR_TOKENS` token (so Ideogram's `Make="Ideogram AI"` is scrubbed too), while keeping genuine camera/editor EXIF. The shared `_is_xai_signature_pair` helper (module-level compiled regexes) is the single source of truth for the pattern, used by both `xai_signature` and `_scrub_ai_exif`. (AVIF/HEIF/JXL still strip only C2PA boxes via `isobmff`, not EXIF — unchanged.)
+**Stripped on removal too:** `remove_ai_metadata` calls `_scrub_ai_exif` on
+JPEG EXIF, which deletes the xAI Signature and UUID Artist pair plus supported
+AI generator values while retaining unrelated camera and editor EXIF. The
+shared `_is_xai_signature_pair` helper is the single source of truth for the
+pair. On the ISOBMFF path, `blank_ai_exif_tokens` provides the corresponding
+in-place scrub for supported EXIF values, TC260 AIGC blocks, and the xAI pair.
 - **China TC260 AIGC label (caught by `AIGC_MARKERS` / `metadata.aigc_label`, surfaced by `identify` as the `aigc` signal):** China-served generators embed an XMP `<TC260:AIGC>{"Label":"1","ContentProducer":...}` block — China's mandatory AI-content labeling (TC260 namespace `tc260.org.cn/ns/AIGC`).
 
-**Doubao** (ByteDance) uses it (verified on the real #13 sample 2026-05-25; `ContentProducer` `001191110102MACQD9K64010000`, no C2PA/SynthID/imwatermark — the XMP block is the only signal; GitHub attachment upload did NOT strip it). The same standard is mandatory for Jimeng/Kling/Qwen/Ernie etc., so the one marker covers the whole China-AIGC-labeled ecosystem. `aigc_label` reads **four serializations** through a shared `_parse` helper: the HTML-entity-encoded XMP `TC260:AIGC` block in **either RDF form** — the nested element `<TC260:AIGC>{...}</TC260:AIGC>` (Doubao) or the attribute `TC260:AIGC="{...}"` (**PicWish**, `ContentProducer="picwish"`, verified on the corpus 2026-05-30) — via a container-agnostic raw-byte scan (any JSON object accepted), a raw-JSON PNG `AIGC` tEXt chunk (Doubao also writes the label this way, no namespaced marker at all — confirmed on the corpus 2026-05-28, `ContentProducer="doubao"`), a bare raw-JSON `{"AIGC":{...}}` object embedded in **JPEG EXIF (UserComment)** by some China-served generators, brace-matched from the scan head with `json.JSONDecoder().raw_decode` (no namespaced marker, no PNG chunk — confirmed on the corpus 2026-05-30, `ContentProducer="001191440300708461136T1308L"`), **and** a bare `AIGC{...}` blob (the label glued straight to its JSON, no `"AIGC":` key wrapper) embedded in a **JPEG APP segment near the JFIF header** — confirmed on the corpus 2026-06-10 (`ContentProducer="00119144030008867405X210002"`; 3 files read `unknown` before this form was added). The two raw-JSON forms are scanned in one loop (`'"AIGC"'` then `AIGC{`) that **falls through on a non-TC260 / undecodable hit instead of returning** — a quoted `"AIGC"` can appear later in an XMP packet while the real label is a bare `AIGC{...}` earlier in the file, so an unconditional early return on the quoted form would shadow the bare form (the exact bug behind the 06-10 misses). All three generic forms (the PNG chunk, the bare `{"AIGC":...}` object, and the bare `AIGC{...}` blob) are gated on at least one TC260 field (`_TC260_FIELDS`) so a generic `AIGC` key cannot false-positive; the namespaced XMP element is unambiguous and needs no gate. `_TC260_FIELDS` covers **two schemas**: the producer-side one (`Label` / `ContentProducer` / `ProduceID` / `ContentPropagator` / `PropagateID`, Doubao and most China gens) and the **service-provider** one (`ServiceProvider` / `ServiceUser`, plus generic `Time` / `ContentId` which are NOT gated on) — **Tencent Cloud's** AIGC variant (`ServiceProvider` = `腾讯云`), embedded in **EXIF `ImageDescription`**, mined from the retained corpus 2026-07 (11 files read `unknown` before — the block was found by the raw-JSON scan but rejected because none of its fields were in the producer-only gate; removal already stripped it since it lives in EXIF). In `identify`, `aigc` fires on the parsed label **or** the `AIGC_MARKERS` byte scan (the latter preserves the laundering-tell case where the JSON payload is truncated).
-- **HuggingFace-hosted job (caught by `metadata.huggingface_job`, surfaced by `identify` as the `hf_job` signal, MEDIUM confidence):** HuggingFace Jobs / Spaces stamp generated PNGs with an `hf-job-id` tEXt chunk holding the job UUID (3 on the corpus 2026-05-28, no other signal). It marks the *hosting job*, not a model — most commonly diffusion output — so it lifts an Unknown verdict to a tentative AI via `hf_only` (parallel to the visible sparkle) but never overrides a hard metadata signal; `_HF_JOB_CAVEAT` states the limit (job, not model; not proof of AI pixels). Stripped on removal (the PNG save whitelist keeps only `STANDARD_METADATA_KEYS`, so `hf-job-id` and the `AIGC` chunk are both dropped). The exact writer is not authoritatively documented (HF Jobs are generic GPU jobs), hence medium not high.
-- **No detectable signal on download (correctly reported `unknown`):** **Recraft** (PNG export is a re-encoded design export — strips everything), **Krea hosting FLUX 2** (no imwatermark despite FLUX — the host omits the encoder, same as Stability's hosted SDXL), and Midjourney (embeds nothing). Lesson: the imwatermark detector only fires on *pristine* output from a pipeline that runs the encoder (diffusers default, official BFL), not from re-hosts (Krea/Stability) or re-encoded exports (Recraft/Canva).
+**Doubao** (ByteDance) uses it (verified on a public issue sample; `ContentProducer` `001191110102MACQD9K64010000`, no C2PA/SynthID/imwatermark — the XMP block is the only signal; GitHub attachment upload did NOT strip it). The same standard is mandatory for Jimeng/Kling/Qwen/Ernie etc., so the one marker covers the whole China-AIGC-labeled ecosystem. `aigc_label` reads **four serializations** through a shared `_parse` helper: the HTML-entity-encoded XMP `TC260:AIGC` block in **either RDF form** — the nested element `<TC260:AIGC>{...}</TC260:AIGC>` (Doubao) or the attribute `TC260:AIGC="{...}"` (**PicWish**, `ContentProducer="picwish"`, verified on compatible samples) — via a container-agnostic raw-byte scan (any JSON object accepted), a raw-JSON PNG `AIGC` tEXt chunk (Doubao also writes the label this way, no namespaced marker at all — confirmed on compatible samples, `ContentProducer="doubao"`), a bare raw-JSON `{"AIGC":{...}}` object embedded in **JPEG EXIF (UserComment)** by some China-served generators, brace-matched from the scan head with `json.JSONDecoder().raw_decode` (no namespaced marker, no PNG chunk — confirmed on compatible samples, `ContentProducer="001191440300708461136T1308L"`), **and** a bare `AIGC{...}` blob (the label glued straight to its JSON, no `"AIGC":` key wrapper) embedded in a **JPEG APP segment near the JFIF header** — confirmed on compatible samples. The two raw-JSON forms are scanned in one loop (`'"AIGC"'` then `AIGC{`) that **falls through on a non-TC260 / undecodable hit instead of returning** — a quoted `"AIGC"` can appear later in an XMP packet while the real label is a bare `AIGC{...}` earlier in the file, so an unconditional early return on the quoted form would shadow the bare form (the exact bug behind the 06-10 misses). All three generic forms (the PNG chunk, the bare `{"AIGC":...}` object, and the bare `AIGC{...}` blob) are gated on at least one TC260 field (`_TC260_FIELDS`) so a generic `AIGC` key cannot false-positive; the namespaced XMP element is unambiguous and needs no gate. `_TC260_FIELDS` covers **two schemas**: the producer-side one (`Label` / `ContentProducer` / `ProduceID` / `ContentPropagator` / `PropagateID`, Doubao and most China gens) and the **service-provider** one (`ServiceProvider` / `ServiceUser`, plus generic `Time` / `ContentId` which are NOT gated on) — **Tencent Cloud's** AIGC variant (`ServiceProvider` = `腾讯云`), embedded in **EXIF `ImageDescription`**, verified on compatible samples. In `identify`, `aigc` fires on the parsed label **or** the `AIGC_MARKERS` byte scan (the latter preserves the laundering-tell case where the JSON payload is truncated).
+- **HuggingFace-hosted job (caught by `metadata.huggingface_job`, surfaced by `identify` as the `hf_job` signal, MEDIUM confidence):** HuggingFace Jobs / Spaces can stamp generated PNGs with an `hf-job-id` tEXt chunk holding the job UUID. It marks the *hosting job*, not a model, so it lifts an Unknown verdict to a tentative AI via `hf_only` but never overrides a hard metadata signal. `_HF_JOB_CAVEAT` states the limit. Removal drops the chunk through the PNG metadata whitelist.
+- **No detectable signal on some downloads:** Recraft exports and some hosted
+  FLUX surfaces can arrive without a supported local signal. Midjourney samples
+  may carry IPTC metadata but no registered C2PA or pixel watermark. The open
+  DWT-DCT decoder only applies when the producing pipeline actually ran its
+  encoder and the carrier remains decodable.
 - **Invisible but NOT locally detectable (proprietary, API/oracle only — same wall as SynthID):** Amazon Titan Image Generator + Nova Canvas (Bedrock `DetectGeneratedContent` API), Kakao (new SynthID image adopter, May 2026), NVIDIA Cosmos (SynthID video). No local detector possible; treat like SynthID.
 - **C2PA 2.4 "Durable Content Credentials" (April 2026; verified against the spec) raise the bar for metadata stripping.** 2.4 defines soft bindings (an invisible watermark or a content fingerprint) plus a server-side manifest repository and a new `c2pa.repository-receipt` assertion. Per the spec: "if a C2PA manifest is removed from an asset, but a copy of that manifest remains in a provenance store elsewhere, the manifest and asset may be matched using available soft bindings." So our local `metadata --remove` deletes the *embedded* manifest, but a fingerprint/watermark soft binding can still re-link the image to its manifest in a repository server-side. Stripping the file is becoming necessary-but-not-sufficient against durable provenance. (Our parsers target the stable embedded-manifest format documented in C2PA 2.1 §11; that format is unchanged in 2.4 -- the new pieces are repository/soft-binding infra, not the on-file box layout, so no parser change is implied.) Spec: https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html We now READ the soft-binding `alg` (`C2PA_SOFT_BINDINGS` / `soft_binding_vendors_in`) to name the forensic-watermark vendor, and locally DECODE the one open scheme, Adobe TrustMark (`trustmark_detector`); the rest (Digimarc/Imatag/Steg.AI/...) stay name-only (proprietary decoders).
-- **Built 2026-05-26 (this batch):** soft-binding `alg` vendor detection; IPTC Photo Metadata 2025.1 AI-disclosure fields (`AISystemUsed` etc.); **video C2PA metadata** detect + strip for MP4/MOV/M4V (free — `isobmff.py` is format-agnostic, MP4 is ISOBMFF); Adobe TrustMark open decoder. NOT done (out of cheap reach, per the feasibility review): visible video-logo removal (needs a video frame pipeline) and audio (SynthID/ElevenLabs/Resemble/Suno all oracle-only or unmarked).
+- **Built in the dated batch:** soft-binding vendor detection, IPTC Photo
+  Metadata AI-disclosure fields, C2PA detection and stripping for supported
+  ISOBMFF video, and the optional Adobe TrustMark decoder. Visible video-logo
+  removal and proprietary audio-watermark detection remain outside the package.
+  Metadata stripping for supported audio containers is a separate implemented
+  path.
 
 **Box detection window — now handled (v0.6.8):** detection no longer relies on a fixed first-MB read. `metadata.scan_head(path, size)` reads the first `size` bytes and, for ISOBMFF, appends the payloads of late provenance boxes found by `isobmff.scan_c2pa_region` (a file-seeking top-level box walker that skips past `mdat` by size without reading it), so a C2PA/AIGC/IPTC manifest placed AFTER a large `mdat` in a streaming/non-faststart MP4 is now caught. Every C2PA/marker byte scan (`has_ai_metadata`, `aigc_label`, `iptc_ai_system`, `synthid_source`, `exif_generator` XMP, `get_ai_metadata` soft-binding, and `identify`) goes through `scan_head`; it is behavior-neutral for non-ISOBMFF inputs (exactly `f.read(size)`).
 
-**Meta-box XMP removal — now handled (v0.6.9):** an AI-label XMP packet stored as a meta-box `mime` item (HEIF/AVIF; out of reach of the top-level box stripper) is blanked in place by `isobmff.blank_ai_xmp_packets` — it locates the packet by its `<?xpacket begin … end?>` delimiters and, if it carries an AI marker (`_AI_LABEL_MARKERS`), overwrites it with spaces of the SAME length, so box sizes / `iloc` offsets stay valid and the coded image is untouched (selective: plain non-AI XMP is left alone, mirroring the top-level uuid logic). Wired into `remove_ai_metadata`'s ISOBMFF branch after `strip_c2pa_boxes`. The remaining gap is an `Exif` meta-box *item* (rare; the AI labels are XMP) — still needs `iinf`/`iloc` surgery or exiftool.
-- **Regulatory driver (context, not a code change):** AI-content labeling mandates are expanding, which pushes more generators toward exactly the C2PA + watermark signals we read. The full per-jurisdiction table lives in README "## Legal" -- keep it there, not duplicated here. Newly added + primary-source verified 2026-05-26: **EU AI Act Article 50** machine-readable marking applicable **2026-08-02** (verified against the article text); **South Korea AI Framework Act Art. 31(3)** in force since **22 January 2026** (verified via Kim & Chang + FPF/Korea Times; Enforcement Decree accepts an invisible-watermark label); **California AB 853** (amends the CA AI Transparency Act) latent-disclosure duty operative **2026-08-02**, requiring a disclosure "permanent or extraordinarily difficult to remove" (verified against the leginfo bill text -- this is the exact disclosure our tool strips); **India IT Amendment Rules 2026** in force **2026-02-20** (verified via Chambers), which prominently-label + permanent-provenance-id all synthetic media AND **expressly prohibit removing/suppressing the label or metadata** -- the first major all-content removal ban outside China.
+**Meta-box XMP and EXIF removal are handled in place:** an AI-label XMP packet
+stored as a meta-box `mime` item is blanked by
+`isobmff.blank_ai_xmp_packets`. Supported EXIF items are handled by
+`blank_ai_exif_tokens`. Both paths preserve box sizes and coded media offsets.
 
-**Removal liability (README "## Legal" disclaimer):** the tool is lawful general-purpose software; liability sits with the remover and is intent-gated -- downstream acts (fraud/deception/IP), plus US DMCA 17 USC 1202 (removing copyright-management info to conceal infringement), plus the removal-as-such bans in China + India. When extending the README table, verify each date/article against the statute/bill text before committing, not against search summaries.
+For current scope and legal context, see
+[scope, safety, and legal notes](legal-and-safety.md). Re-verify legal facts
+against primary sources before adding jurisdiction-specific claims.
 
 ## Visible AI-generation marks + detection methods (deep-research 2026-07-10, adversarially verified)
 
@@ -51,11 +73,14 @@ Grok JPEG downloads (Aurora model) carry **no C2PA, no XMP, no SynthID, no IPTC*
 
 **Visible-mark landscape beyond the registry.** Meta stamps a visible "Imagined with AI" mark (bottom-LEFT, a small symbol) on its OWN Meta AI / "Imagine" output; for third-party images it relies on C2PA / IPTC, not a visible mark. Samsung Galaxy AI additionally uses a **four-star icon** variant in a corner alongside the localized text wordmark `samsung_engine` calibrates (only the Italian text variant is covered) -- the icon is a distinct, uncovered variant. Every source agrees visible + metadata marks are trivially removable (crop / screenshot, ~2 s), which is the tool's premise.
 
-**Regulatory driver -- China GB 45438-2025 is the strongest VISIBLE-mark mandate.** The CAC / TC260 "Measures for Labeling AI-Generated Synthesized Content" (issued March 2025, **effective 2025-09-01**, technical standard **GB 45438-2025**, building on the TC260 Aug-2023 practice guide) MANDATE a **visible** label for AI images -- a visible textual mark whose height must be **>= 5% of the image's shortest side** -- plus the metadata (implicit) label. So every major Chinese platform now ships visible "AI生成"-style text marks (we cover Doubao / Jimeng; expect more CJK-text marks under this driver). By contrast EU AI Act Article 50 mandates only the MACHINE-READABLE mark (enforceable 2026-08-02, grace to 2026-12-02); a visible label is proposed and modality-specific (visible for images) but is NOT a hard "fixed icon" mandate -- a claim that Art 50 requires a clearly-visible fixed icon for images was refuted in verification. Primary-source dates verified against the article/standard text, not search summaries.
+**Regulatory driver -- China GB 45438-2025 is the strongest VISIBLE-mark mandate.** The CAC / TC260 "Measures for Labeling AI-Generated Synthesized Content" (issued March 2025, **effective 2025-09-01**, technical standard **GB 45438-2025**, building on the TC260 Aug-2023 practice guide) MANDATE a **visible** label for AI images -- a visible textual mark whose height must be **>= 5% of the image's shortest side** -- plus the metadata (implicit) label. Several such CJK text marks are now registered; see [supported signals](supported-signals.md) for the current list. By contrast EU AI Act Article 50 mandates only the MACHINE-READABLE mark (enforceable 2026-08-02, grace to 2026-12-02); a visible label is proposed and modality-specific (visible for images) but is NOT a hard "fixed icon" mandate -- a claim that Art 50 requires a clearly-visible fixed icon for images was refuted in verification. Primary-source dates verified against the article/standard text, not search summaries.
 
 ## Uncovered visible marks: implementation specs (deep-research 2026-07-18)
 
-Triggered by a corpus finding: **50% of TC260-labelled uploads (1452 of 2896 unique) produced no detection at all**, and hand-inspection showed the bulk are MISSED Doubao marks (a localization defect, now fixed -- see `scale_basis` in `docs/module-internals.md`) plus a minority of genuinely uncovered vendors. Verification status is labelled per claim; treat (b)/(c) as leads, not ground truth.
+Compatibility testing showed that TC260-labelled images can still produce no visible-mark
+detection. The main causes were a fixed Doubao localization defect and genuinely
+uncovered vendors. Verification status is labelled per claim; treat (b)/(c) as leads,
+not ground truth.
 
 **GB 45438-2025 clause 5.2, the binding constraint for every Chinese mark (VERIFIED (a) -- full standard text extracted from the TC260-hosted PDF).** Verbatim requirements for an image's explicit label:
 - 应采用文字提示 (must be a TEXT prompt);
@@ -72,12 +97,12 @@ Two consequences we can exploit: (1) the 5% floor is a **scale prior** -- a comp
 
 **Baidu: RESOLVED 2026-07-22, registered (`baidu_engine.py`).** The mark is a white bold "百度" text run + a separate white rounded tag with dark "AI生成", bottom-right -- settled by the TC260 USCC cohort harvest (16 frames, USCC 91110000802100433B), not by web research. Detection keys on the text run only; details in `docs/module-internals.md`.
 
-**Tencent Yuanbao: RESOLVED 2026-07-25, registered (`yuanbao_engine.py`).** The standard mark is a compact two-line italic `元宝` over `AI生成` block at bottom-right. It switches between light and dark strokes with the scene, so detection uses polarity-independent local contrast rather than a white top-hat. The corrected synthetic silhouette and corpus calibration are recorded in `docs/module-internals.md`; a separate one-line photographer-overlay variant remains evidence-limited to one example.
+**Tencent Yuanbao: RESOLVED 2026-07-25, registered (`yuanbao_engine.py`).** The standard mark is a compact two-line italic `元宝` over `AI生成` block at bottom-right. It switches between light and dark strokes with the scene, so detection uses polarity-independent local contrast rather than a white top-hat. The separate one-line overlay variant remains evidence-limited to one example.
 
-**Meta `Imagined with AI` (string VERIFIED (a) from Meta's own newsroom; POSITION NOT VERIFIED).** Sources conflict (bottom-left vs bottom-right) and one claims newer Meta models dropped the visible mark for invisible watermarking; none survived a fetch. Do NOT encode a corner without a corpus sample. Meta also embeds IPTC + invisible watermarks, which `identify` already reads. Source: `https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/`.
+**Meta `Imagined with AI` (string VERIFIED (a) from Meta's own newsroom; POSITION NOT VERIFIED).** Sources conflict on placement. Do not encode a corner without a verified sample. `identify` reads the supported IPTC disclosure; it does not decode Meta's proprietary invisible watermark. Source: `https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/`.
 
 **Samsung English/other locales: still not established.** Samsung's own support page says only that "A Galaxy AI watermark will appear on AI-generated images" -- no string, no corner. Every community thread carrying the exact English string returned HTTP 403 to WebFetch, so the search paraphrase (bottom-left) is deliberately NOT recorded as fact. Feature-tier detail (b): the mark is applied by Generative Edit / sketch-to-image but reportedly NOT by Object Eraser, so Samsung absence is feature-dependent. The four-star icon variant: nothing found.
 
 **The one document that would settle ByteDance placement is BLOCKED.** Douyin's 《抖音关于人工智能生成内容标识的水印与元数据规范》 aims to give AI tools a unified watermark style and position, which would cover Doubao / Jimeng / 星绘 at once. Both mirrors return HTTP 403 to WebFetch; a secondary report (b, unconfirmed) says the watermark is `AI生成` + tool name + company name placed **top-left** -- which would explain the Jimeng pill's top-left position but contradicts the GB annex's bottom-right example. Worth one retry through Chrome MCP with a real browser session.
 
-**No vendor publishes typeface, colour, opacity, plate, or margin for ANY of these marks.** The only font-adjacent requirement anywhere is GB's "legible typeface". So each synthetic silhouette's font must be calibrated against corpus positives exactly as the Jimeng pill was; candidate CJK families by platform convention (inference): HarmonyOS Sans / Source Han Sans / Noto Sans CJK SC for Android-origin apps, PingFang SC for iOS-origin.
+**No vendor publishes typeface, color, opacity, plate, or margin for ANY of these marks.** The only font-adjacent requirement anywhere is GB's "legible typeface". So each synthetic silhouette's font must be calibrated against corpus positives exactly as the Jimeng pill was; candidate CJK families by platform convention (inference): HarmonyOS Sans / Source Han Sans / Noto Sans CJK SC for Android-origin apps, PingFang SC for iOS-origin.