feat(invisible): protect text automatically by default (#21)

Mirror protect_faces: protect_text defaults to True in invisible_engine and
watermark_remover, so the SDXL pipeline detects text per image and switches to
Differential Diffusion only when glyphs are found. Text-free inputs fall back to
plain img2img with no differential-pipeline load, so the autonomy is free. The
CLI now exposes a single off-switch --no-protect-text instead of the positive
flag, keeping the interface minimal.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

CLAUDE.md

@@ -29,14 +29,16 @@ You are a **principal Python engineer** maintaining a CLI tool and library for r
 
 - `noai/c2pa.py` — PNG chunk parser; use `extract_c2pa_chunk(path)` to get raw caBX payload, `has_c2pa_metadata(path)` to detect. Do not reimplement chunk parsing. `extract_c2pa_info(path)` sets `synthid_watermark`/`synthid_vendors` when the manifest is signed by a SynthID-using vendor, and `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 the PNG parser and the non-PNG binary path).
 - `noai/constants.py` — PNG_SIGNATURE, C2PA_CHUNK_TYPE, C2PA_SIGNATURES, C2PA_ISSUERS, `SYNTHID_C2PA_ISSUERS` (issuers that pair SynthID with C2PA: Google, OpenAI), and `C2PA_SOFT_BINDINGS` (soft-binding `alg` prefix → forensic-watermark vendor: Adobe TrustMark, Digimarc, Imatag, Steg.AI, Microsoft, ...). Add a new issuer/binding here, not inline.
-- `metadata.py` — `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.
-- `identify.py` — `identify(path)` aggregates every locally-readable signal (C2PA issuer→platform, C2PA soft-binding forensic-watermark vendor, IPTC "Made with AI" + IPTC 2025.1 `AISystemUsed`, embedded SD/ComfyUI params, SynthID proxy, xAI/Grok EXIF signature via `metadata.xai_signature`, visible Gemini sparkle, 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). Visible-sparkle is promoted only at confidence ≥ `_SPARKLE_THRESHOLD` (0.5; corpus-tuned to separate Gemini sparkles ≥0.56 from non-sparkle ≤0.49). The cv2 dependency lives in `gemini_engine.detect_sparkle_confidence`, not here. **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`). **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. **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/Samsung/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. 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). **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. 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 device tokens to `_DEVICE_C2PA_PLATFORM`, generator/issuer platforms to `_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` and XMP `CreatorTool` generator tags 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. **Ideogram tags its output with EXIF `Make="Ideogram AI"`** (verified on a real download 2026-05-24) — that's why `Make` is read.
+- `metadata.py` — `scan_head(path, size=1MB)` is the shared input for every C2PA/AIGC/IPTC byte scan: first `size` bytes plus, for ISOBMFF, the late provenance-box payloads from `isobmff.scan_c2pa_region` (catches a manifest after a large `mdat`); behavior-neutral (`f.read(size)`) for non-ISOBMFF. Use it instead of `open().read(1MB)` for any new marker scan. `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.
+- `identify.py` — `identify(path)` aggregates every locally-readable signal (C2PA issuer→platform, C2PA soft-binding forensic-watermark vendor, IPTC "Made with AI" + IPTC 2025.1 `AISystemUsed`, embedded SD/ComfyUI params, SynthID proxy, xAI/Grok EXIF signature via `metadata.xai_signature`, visible Gemini sparkle, 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). Visible-sparkle is promoted only at confidence ≥ `_SPARKLE_THRESHOLD` (0.5; corpus-tuned to separate Gemini sparkles ≥0.56 from non-sparkle ≤0.49). The cv2 dependency lives in `gemini_engine.detect_sparkle_confidence`, not here. **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`). **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. **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/Samsung/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. 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). **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. 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 device tokens to `_DEVICE_C2PA_PLATFORM`, generator/issuer platforms to `_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` and XMP `CreatorTool` generator tags 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. **Ideogram tags its output with EXIF `Make="Ideogram AI"`** (verified on a real download 2026-05-24) — that's why `Make` is read. **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 any AI-generation marker. Vendor normalizatLine truncated
 - `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`.
 - `doubao_engine.py` — visible Doubao "豆包AI生成" remover/detector (cv2/numpy, no GPU). `DoubaoEngine.locate` anchors a bottom-right box by **geometry** (mark scales with image WIDTH, fractions in module constants; no bundled template), `extract_mask` pulls the light low-saturation glyphs with a **polarity-aware white top-hat** (brighter-than-blurred-local-bg, so white-paper documents are left untouched instead of smeared), `detect` thresholds glyph coverage (`DETECT_MIN_COVERAGE` 0.16 separates real marks ≥0.20 from corner noise, which stays ≤0.06 on large images but can spike to ~0.15 on tiny ones), `remove_watermark` inpaints (cv2 Telea/NS) and **bails when coverage > `MAX_INPAINT_COVERAGE` 0.50** (dense-text background → would smear). Wired into `visible --mark` via `cli._run_doubao_if_selected`. **Logo is near-white (~253), not the gray some third-party tools assume.** Best on photo/illustration backgrounds; high-contrast edges leave faint residue (cv2-inpaint limit). Clean per-pixel reverse-alpha (Gemini-style) is the future upgrade but needs a captured/distilled alpha map — see below.
 - `region_eraser.py` — universal region eraser (`erase` CLI). `erase(image, boxes=|mask=, backend=)`: `boxes_to_mask` → `cv2.inpaint` (`cv2` backend, default, no deps) or big-LaMa via onnxruntime (`lama` backend, extra `lama`, `Carve/LaMa-ONNX` Apache-2.0 model downloaded on first use, never bundled). `erase_lama` crops a padded region around the mask, runs LaMa at its fixed 512² input, pastes only masked pixels back (untouched areas stay pixel-exact). Lazy `_get_lama_session` singleton; `lama_available()` guards the optional import. **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` — `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` — `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, 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.
+- `text_protector.py` — text-region protection for the `invisible` SDXL img2img pass (issue #21: CJK/small text deforms at watermark-removal strengths). `is_available()` gates on `cv2.dnn.TextDetectionModel_DB`; `TextProtector.detect_text_boxes(bgr)` runs the **PP-OCRv3 CN** ONNX detector (~2.4 MB, Apache-2.0, opencv_zoo, CJK-native, returns rotated quad polygons) — downloaded+cached to `~/.cache/remove-ai-watermarks` on first use via atomic temp-rename, never bundled, **no torch (cv2.dnn only)**. `build_change_map(boxes, h, w, preserve=0.9, feather=15)` paints a Differential-Diffusion change map. **Polarity (verified empirically):** white(1.0)=PRESERVE original pixels, black(0.0)=MAX change; map is black bg + `preserve` inside text polygons, Gaussian-feathered edges, clipped to [0,1]. `preserve` stays below a hard 1.0 freeze by default so text still scrubs lightly (SynthID survives cropping). Wired into `watermark_remover._run_differential` via the community `pipeline_stable_diffusion_xl_differential_img2img` (loaded with `custom_revision="0.38.0"` — HF resolves the **PyPI** version string, not the `v0.38.0` git tag); gated to the SDXL `DEFAULT_MODEL_ID` only (`_can_protect_text`), falls back to plain img2img otherwise. **Autonomous by default** (`protect_text=True` in `invisible_engine`/`watermark_remover`, mirroring `protect_faces`): the detector runs per image and `_run_differential` falls back to plain img2img when **no boxes** are found, so text-free inputs pay only the cheap cv2 detection (no differential-pipeline load). CLI exposes a single off-switch `--no-protect-text` on `invisible`/`all` (passed as `protect_text=not no_protect_text`); the unavailable-model case logs at debug, not warning, since it is now the default path. The diff pipeline upcasts the VAE to fp32 internally, so do **not** add `upcast_vae()`/`enable_attention_slicing` (both produced NaN/black on fp16 MPS). `build_change_map` is unit-tested without any model download (`tests/test_text_protector.py`).
 - `face_protector.py` — YOLO detect + soft-blend pattern; mirror this for any "protect region during diffusion" features
+- `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.** macOS/Linux already accept UTF-8 paths, so it is behavior-neutral there (the bug only reproduces on Windows). cv2/numpy are imported lazily inside the functions, so the module is cheap to import in a bare env.
 
 ### Doubao clean-reverse-alpha distillation (researched 2026-05-26, NOT yet shipped)
 
@@ -53,15 +55,16 @@ Who embeds what, and whether it is locally detectable (so we know which gaps are
 - **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).
 - **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). The soft-binding **box detection window**: non-PNG/video detection scans the first 1 MB, so a C2PA box placed after a large `mdat` in a streaming MP4 can be missed — front-placed manifests (the common case) are caught.
-- **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 Jan 2026 (verified via Kim & Chang; 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. **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.
+- **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). **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. **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.
 
 ## Known limitations
 
 - `invisible` pipeline processes at **native resolution by default** (`max_resolution=0`), matching the hosted raiw.cc backend (fal fast-sdxl, no pre-downscale). The old forced downscale-to-1024 -> upscale-back round-trip was the main quality loss (issue #10) and is gone; at strength ~0.05 SDXL img2img does not need the ~1024 downscale. `--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-based diffusion is still the proper long-term fix. **Concrete MPS data point (verified 2026-05-25 on a 1254x1254 gpt-image SDXL run, fp32, 20 GB MPS ceiling):** 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. Adding `enable_vae_tiling()` alone does NOT prevent it (the peak is the UNet, not the VAE). The fast Mac workarounds are fp16 on MPS (roughly halves memory) or `--max-resolution` to cap the long side; neither is wired as the default. The native-vs-downscale decision lives in the pure helper `invisible_engine._target_size(w, h, max_resolution)` (returns `None` for native, a clamped target tuple otherwise) so it is unit-tested (`tests/test_invisible_engine.py::TestTargetSize`, the #10/#15 regression guard) without loading the model -- keep that logic in the helper, don't re-inline it.
 - Pyright first run is slow (2-3 min) due to ML deps (torch/diffusers/transformers stubs); full-project `uv run pyright` can stall for many minutes — scope it to changed files.
 - `ultralytics` monkey-patches `PIL.Image.open` and tries to autoload `pi_heif`. When `pi_heif` is missing, opening files raises `ModuleNotFoundError`, not `UnidentifiedImageError`. Code that opens user-supplied or unknown-format files should `except Exception`, not just `OSError`/`UnidentifiedImageError`.
-- 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. **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). **Still NOT built (deliberate):** EXIF/XMP stored as *items inside the ``meta`` box* (typical for AVIF/HEIF images) needs meta-box surgery (iinf/iloc edit + mdat splice) with corruption risk -- exiftool would do it but is a non-installed binary dep, so it stays a documented gap. **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.)
+- **rich `console.print` parses `[word]` as a style tag and silently drops unknown ones.** A literal bracketed token in a print string disappears: `pip install 'remove-ai-watermarks[gpu]'` rendered as `...remove-ai-watermarks'` (the `[gpu]` extra eaten), which sent users a broken install command (surfaced via #19). Escape the literal bracket as `\[gpu]` (in a normal Python string that is `"\\[gpu]"`) in any rich string carrying user-facing brackets. Regression-guarded by `tests/test_cli.py::TestGpuHintMarkup`.
+- 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. **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). **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). **Still NOT built:** an `Exif` *item* inside the `meta` box (rare -- AI labels are XMP) needs full `iinf`/`iloc` surgery (offset rewrite) with corruption risk -- exiftool (R/W/C for HEIC/AVIF EXIF+XMP, verified on exiftool.org 2026-05-27) would do it but is a non-installed binary dep, so it stays a documented gap. **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.)
 - **SynthID detection is metadata-only.** 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. **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). 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. **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.) **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 deteLine truncated
 - **External AI-vs-real classifier models are out of scope (decided 2026-05-24).** 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.
 - **SynthID v2 vs default pipeline:** the SDXL-based default profile (since May 2026) defeats SynthID v2. **Verified end-to-end (May 2026):** local SDXL run on a Gemini 3 Pro output, checked via the Gemini app's "Verify with SynthID" feature, returned "no SynthID watermark detected". Also confirmed against **OpenAI's** SynthID (2026-05-23): a fresh ChatGPT/gpt-image output read "SynthID detected" on openai.com/verify before the local SDXL run and "SynthID not detected" after (corpus regression chain: pos `4ef377bd` -> cleaned `47188e88`). The same configuration is used in raiw-app production (`fal-ai/fast-sdxl/image-to-image`, strength 0.05, steps 50, guidance 7.5, no pre-downscale). fal's own `llms.txt` for `fast-sdxl` names the base checkpoint as `stabilityai/stable-diffusion-xl-base-1.0` (verified 2026-05-25) -- the exact checkpoint the local CLI defaults to (`DEFAULT_MODEL_ID`). So the local `invisible` default is weight-for-weight identical to prod; "fast-sdxl" is fal's optimized serving, not different weights. After the native-resolution fix the local pipeline matches prod on weights + strength + steps + guidance + resolution. SD-1.5 dreamshaper at 768 px was previously the default and does NOT defeat v2 — verified empirically against the same feature (strength 0.04, 0.10, and elastic warp α∈{5,8} all flagged positive). That SD-1.5 path was removed; only `default` (SDXL) and `ctrlregen` profiles remain. **Scope of the claim: defeating the SynthID verifier is NOT the same as forensic invisibility.** "Removing the Watermark Is Not Enough: Forensic Stealth in Generative-AI Watermark Removal" (arXiv:2605.09203, 2026-05) shows that six removal attacks across four families (UnMarker, CtrlRegen+, WatermarkAttacker, etc.) all leave forensic traces: independent detectors flag *removal-processed* images vs genuinely-clean ones at **>98% TPR at 1% FPR**. So our SDXL pass makes the oracle read "SynthID not detected," but the output can still be classifiable as "an image that went through a removal pipeline." Do not over-claim "indistinguishable from a real photo." https://arxiv.org/abs/2605.09203

README.md

@@ -12,14 +12,16 @@ Strips SynthID, C2PA Content Credentials, EXIF/XMP "Made with AI" labels, and vi
 
 If this tool saves you time, consider [sponsoring its development](https://github.com/sponsors/wiltodelta).
 
+> **Intended for lawful use only.** Publishing and running this software is lawful; responsibility for any downstream use, and for compliance with local law, rests entirely with the user. Some jurisdictions restrict removing an AI label as such (see [Legal](#legal)). The authors do not condone use for deception, fraud, or any unlawful activity.
+
 ## Features
 
 - **Visible watermark removal** — Gemini / Nano Banana sparkle logo (reverse alpha blending) and the Doubao "豆包AI生成" text strip (locate + mask + inpaint); fast, offline, deterministic, no GPU. `visible --mark auto` picks the right one
 - **Universal region eraser (`erase`)** — remove any logo / watermark / object inside boxes you specify, regardless of position or colour. Default cv2 inpainting (CPU, instant); optional big-LaMa via onnxruntime (`lama` extra) for higher quality
 - **Invisible watermark removal** — SynthID, StableSignature, TreeRing via diffusion-based regeneration (needs a local GPU, or run it with no setup on [raiw.cc](https://raiw.cc))
-- **AI metadata stripping** — EXIF, PNG text chunks, C2PA provenance manifests (PNG / JPEG / AVIF / HEIF / JPEG-XL, and **MP4 / MOV / M4V video** at the container level), XMP DigitalSourceType
-- **"Made with AI" label removal** — removes the metadata that triggers AI labels on Instagram, Facebook, X (Twitter)
-- **Analog Humanizer** — film grain and chromatic aberration to bypass AI image classifiers
+- **AI metadata stripping** — EXIF, PNG text chunks, C2PA provenance manifests (PNG / JPEG / AVIF / HEIF / JPEG-XL, **MP4 / MOV / M4V / M4A** at the container level, and **WebM / MP3 / WAV / FLAC / OGG** losslessly via ffmpeg), XMP DigitalSourceType
+- **"Made with AI" label removal** — removes the AI-disclosure metadata that platforms read to apply automatic labels (useful for clearing a false-positive label from a human-edited photograph)
+- **Analog Humanizer** — optional film grain and chromatic aberration post-processing
 - **Smart Face Protection** — automatic extraction and blending of human faces to prevent AI distortion
 - **Batch processing** — process entire directories
 - **Detection** — three-stage NCC watermark detection with confidence scoring
@@ -105,6 +107,8 @@ SDXL is the default since May 2026: empirically defeats SynthID v2 on Gemini 3 P
 
 **Analog Humanizer**: optional film grain and chromatic aberration injection that mimics a photo of a screen, raising the bar for AI-generated image classifiers. (It frustrates generic classifiers but does not guarantee forensic invisibility — see the [arXiv:2605.09203](https://arxiv.org/abs/2605.09203) note above.)
 
+**Text Protection** (automatic): SDXL img2img regenerates every pixel, so small text and CJK glyphs get deformed at the strengths that defeat SynthID. The SDXL pipeline guards against this by default: a CJK-native PP-OCRv3 text detector (a 2.4 MB ONNX model run on CPU via OpenCV's DNN module, downloaded and cached on first use) locates text regions, and if any are found the pass switches to Differential Diffusion so a per-pixel change map keeps the text regions largely intact while the background is regenerated normally. Text-free images run the standard pass at no extra cost. Pass `--no-protect-text` to turn it off. SDXL default pipeline only.
+
 ### Stripping C2PA, EXIF, and "Made with AI" metadata
 
 AI tools embed generation metadata that social platforms use to show "Made with AI" labels:
@@ -245,6 +249,7 @@ remove-ai-watermarks erase image.png --region 1640,1930,400,100 -o clean.png
 remove-ai-watermarks invisible image.png -o clean.png --humanize 4.0
 # Runs at native resolution by default. On a very large image that OOMs the
 # GPU/MPS, cap the long side: --max-resolution 2048
+# Text / CJK glyphs are preserved automatically; disable with --no-protect-text
 
 # Check / strip AI metadata (C2PA, EXIF, "Made with AI" labels)
 # --check also flags SynthID-bearing sources: a C2PA manifest signed by
@@ -322,7 +327,9 @@ Tracked but not yet implemented:
 - **Grow the SynthID reference corpus** (`data/synthid_corpus/`) with oracle-labeled samples per model and resolution (Gemini app for Google, openai.com/verify for OpenAI). Prerequisite for any pixel-detector attempt and for an automated removal-regression set.
 - **Real non-PNG C2PA fixtures**. SynthID-source detection for JPEG / WebP / AVIF is currently covered only by synthetic byte blobs; replace with real vendor-emitted files to ground the binary-scan path.
 - **Maintenance debt**. Clear strict-pyright debt in `remove_ai_metadata` / `cli.py` (untyped piexif / PIL / click / rich) so `maintain.sh` can finish green. (`uv-secure` is already clean since `idna` was bumped to 3.16.)
-- **AVIF / HEIF / JPEG-XL detection limits**. Removal strips top-level C2PA `uuid` and JUMBF `jumb` boxes. EXIF/XMP boxes inside these containers are not yet scrubbed (PNG and JPEG are fully covered).
+- **AVIF / HEIF `Exif` item inside the `meta` box**. An AI-label *XMP* packet in a `meta`-box item is now blanked in place (v0.6.9), but EXIF stored as a `meta`-box `Exif` *item* is still not removed — it needs full `iinf`/`iloc` surgery (offset rewrite, corruption risk) or `exiftool` (a non-bundled binary dependency). Low priority: the AI labels we target are XMP, not EXIF, so an EXIF-only meta-box case is rare.
+- **More C2PA device signers**. Leica, Nikon, Google Pixel, Sony, and Truepic are mapped (each verified against a real signed file). Canon and Samsung Galaxy (AI-edit) are deferred until a real signed sample surfaces — no public direct-download C2PA file exists for them today (upload-to-verify / news-agency-licensed only).
+- **Resemble PerTh audio detection** — evaluated, not feasible with the public API: `get_watermark()` returns a raw bit array with no presence/confidence flag, so watermarked vs. clean audio can't be reliably separated without Resemble's fixed payload or a confidence service. Same wall as the SynthID pixel detector.
 - **Video pipeline (`noai-video`)**: per-frame inpainting and tracking for Sora 2 dynamic logo, Veo 3.1 badge, Kling, Runway. Separate package, not folded into this repo.
 
 Won't fix:
@@ -336,10 +343,10 @@ Watermarking and provenance for AI-generated content is now regulated in several
 | Jurisdiction | Instrument | Status (May 2026) | Relevance |
 | --- | --- | --- | --- |
 | EU | AI Act, Article 50 | Transparency duties apply from **2 August 2026**. Legacy generative systems (placed on the market before that date) get a grandfathering period to **2 December 2026** for the Article 50(2) marking duty, under the Digital Omnibus (Commission proposal Nov 2025; co-legislator political agreement 7 May 2026). Article 50 guidelines and a marking Code of Practice are being finalised through 2026. | Removing mandated provenance markers with intent to deceive may be sanctioned under national implementations. |
-| US (federal) | COPIED Act | **Reintroduced April 2025; not enacted** (pending in the Senate). | If passed, would set NIST provenance standards and prohibit tampering with / removing provenance information. The tool itself is lawful; usage may not be. |
-| US (state) | CA AB 2655, TX SB 751, similar | TX SB 751 in force; **CA AB 2655 struck down** by a federal court (Aug 2025, Section 230 / First Amendment). | Content-specific (election deepfakes, sexual deepfakes). Not tool-specific. |
+| US (federal) | COPIED Act (S. 1396, 119th Cong.) | **Reintroduced April 2025; not enacted** (referred to Senate Commerce Committee). | If passed, would set NIST provenance standards and prohibit tampering with / removing provenance information. The tool itself is lawful; usage may not be. |
+| US (state) | CA AB 2655, TX SB 751 (2019), similar | TX SB 751 (2019) in force; **CA AB 2655 struck down** by a federal court (E.D. Cal., Aug 2025, *Kohls v. Bonta*) as preempted by **Section 230**; the court did not reach the First Amendment (the companion law AB 2839 was separately enjoined on First Amendment grounds). | Content-specific (election deepfakes, sexual deepfakes). Not tool-specific. |
 | US (state) | CA AB 853 (amends the California AI Transparency Act) | Core provider duties operative **2 August 2026** (delayed from 1 January 2026); large platforms 1 January 2027; capture devices 1 January 2028. | Covered providers (1M+ monthly users) must embed a latent disclosure that is "permanent or extraordinarily difficult to remove" and offer a free detection tool. Removing that disclosure is what this tool does. |
-| South Korea | AI Framework Act (Basic Act on AI), Article 31 | In force since **January 2026** (one-year transition after promulgation). | Art. 31(3): AI output "difficult to distinguish from reality" must be labeled so users "clearly recognize" it; the draft Enforcement Decree accepts a machine-readable (invisible-watermark) label. Artistic/creative works get a presentation exception. |
+| South Korea | AI Framework Act (Basic Act on AI), Article 31 | In force since **22 January 2026** (one-year transition after promulgation). | Art. 31(3): AI output "difficult to distinguish from reality" must be labeled so users "clearly recognize" it; the draft Enforcement Decree accepts a machine-readable (invisible-watermark) label. Artistic/creative works get a presentation exception. |
 | China | Measures for Labeling AI-Generated Content (+ GB 45438-2025) | In force since **1 September 2025**. | Mandatory explicit (visible) + implicit (metadata) labels across image / audio / video; tampering with, forging, or removing labels is prohibited. |
 | India | IT (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 | In force since **20 February 2026** (notified 10 February 2026). | All "synthetically generated information" must be **prominently labelled** and carry **permanent metadata / a provenance identifier**; the rules expressly **prohibit modifying, suppressing, or removing** that label or metadata. Covers image, audio, and audio-visual content. |
 | UK | Online Safety Act 2023 / Ofcom guidance | In force, but **no statutory AI-provenance or watermarking obligation**. | Ofcom encourages watermarking / provenance metadata as voluntary "attribution measures"; platform duties, not user obligations. |

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "remove-ai-watermarks"
-version = "0.6.5"
+version = "0.6.10"
 description = "Remove visible and invisible AI watermarks from images (Gemini / Nano Banana, ChatGPT, Stable Diffusion)"
 readme = "README.md"
 requires-python = ">=3.10"
@@ -19,7 +19,17 @@ dependencies = [
 gpu = [
     "torch>=2.0.0",
     "diffusers>=0.38.0",
-    "transformers>=4.35.0",
+    # diffusers 0.38's auto-pipeline registry imports ``Qwen3VLForConditional
+    # Generation`` (its ``nucleusmoe_image`` pipeline), which only exists in
+    # transformers 5.x -- so ``from diffusers import AutoPipelineForImage2Image``
+    # fails on transformers 4.x. The real SDXL-loading break was NOT transformers
+    # 5.x but the tokenizers *release candidate* (0.23.0rc0) that the global
+    # ``prerelease = "allow"`` drags in: its CLIP tokenizer raises
+    # ``RobertaProcessing.__new__() got an unexpected keyword argument 'cls'``.
+    # Cap tokenizers to the stable 0.22 line (transformers 5.x accepts
+    # >=0.22,<=0.23.0) so the rc is excluded while SDXL still loads.
+    "transformers>=5,<6",
+    "tokenizers>=0.22,<0.23",
     "accelerate>=0.25.0",
     "controlnet-aux>=0.0.9",
     "safetensors",

src/remove_ai_watermarks/__init__.py

@@ -1,3 +1,3 @@
 """Remove-AI-Watermarks: Unified tool for removing visible and invisible AI watermarks."""
 
-__version__ = "0.6.5"
+__version__ = "0.6.9"

src/remove_ai_watermarks/cli.py

@@ -85,7 +85,9 @@ def _read_bgr_and_alpha(path: Path) -> tuple[np.ndarray | None, np.ndarray | Non
     """
     import cv2
 
-    image = cv2.imread(str(path), cv2.IMREAD_UNCHANGED)
+    from remove_ai_watermarks import image_io
+
+    image = image_io.imread(path, cv2.IMREAD_UNCHANGED)
     if image is None:
         return None, None
     if image.ndim == 2:
@@ -109,11 +111,12 @@ def _write_bgr_with_alpha(
     forced to 0 inside that bbox (expanded by ``pad`` px) so the watermark area
     becomes fully transparent in the saved file.
     """
-    import cv2
     import numpy as np
 
+    from remove_ai_watermarks import image_io
+
     if alpha is None or path.suffix.lower() not in _ALPHA_FORMATS:
-        cv2.imwrite(str(path), bgr)
+        image_io.imwrite(path, bgr)
         return
 
     alpha_out = alpha
@@ -127,7 +130,7 @@ def _write_bgr_with_alpha(
             alpha_out[y0:y1, x0:x1] = 0
 
     bgra = np.dstack([bgr, alpha_out])
-    cv2.imwrite(str(path), bgra)
+    image_io.imwrite(path, bgra)
 
 
 def _run_doubao_if_selected(
@@ -457,6 +460,12 @@ def cmd_erase(
     default=0,
     help="Cap long side (px) before diffusion; 0 = native (best quality, like raiw.cc). Raise only on GPU/MPS OOM.",
 )
+@click.option(
+    "--no-protect-text",
+    is_flag=True,
+    default=False,
+    help="Disable automatic text protection (text/CJK is preserved by default on the SDXL pipeline).",
+)
 @click.pass_context
 def cmd_invisible(
     ctx: click.Context,
@@ -470,6 +479,7 @@ def cmd_invisible(
     hf_token: str | None,
     humanize: float,
     max_resolution: int,
+    no_protect_text: bool,
 ) -> None:
     """Remove invisible AI watermarks (SynthID, StableSignature, TreeRing).
 
@@ -481,7 +491,7 @@ def cmd_invisible(
     if not invisible_available():
         console.print(
             "[red]Error:[/] GPU dependencies not installed.\n"
-            "  Install them with: [bold]pip install 'remove-ai-watermarks[gpu]'[/]"
+            "  Install them with: [bold]pip install 'remove-ai-watermarks\\[gpu]'[/]"
         )
         raise SystemExit(1)
 
@@ -516,6 +526,7 @@ def cmd_invisible(
         guidance_scale=None,
         seed=seed,
         humanize=humanize,
+        protect_text=not no_protect_text,
         max_resolution=max_resolution,
     )
     elapsed = time.monotonic() - t0
@@ -615,6 +626,11 @@ def cmd_identify(ctx: click.Context, source: Path, no_visible: bool, as_json: bo
     console.print(f"\n  Verdict: {verdict}  [dim](confidence: {report.confidence})[/]")
     console.print(f"  Platform: {report.platform or '[dim]undetermined[/]'}")
 
+    if report.integrity_clashes:
+        console.print("\n  [bold red]⚠ Integrity clash[/] [dim](provenance signals contradict each other)[/]")
+        for clash in report.integrity_clashes:
+            console.print(f"  [red]- {clash}[/]")
+
     if report.watermarks:
         table = Table(show_header=True, header_style="bold", title="Watermarks / provenance markers")
         table.add_column("Marker", style="cyan")
@@ -663,6 +679,12 @@ def cmd_identify(ctx: click.Context, source: Path, no_visible: bool, as_json: bo
     default=0,
     help="Cap long side (px) before diffusion; 0 = native (best quality, like raiw.cc). Raise only on GPU/MPS OOM.",
 )
+@click.option(
+    "--no-protect-text",
+    is_flag=True,
+    default=False,
+    help="Disable automatic text protection (text/CJK is preserved by default on the SDXL pipeline).",
+)
 @click.pass_context
 def cmd_all(
     ctx: click.Context,
@@ -679,6 +701,7 @@ def cmd_all(
     hf_token: str | None,
     humanize: float,
     max_resolution: int,
+    no_protect_text: bool,
 ) -> None:
     """Remove ALL watermarks: visible + invisible + metadata.
 
@@ -744,7 +767,7 @@ def cmd_all(
         if not invisible_available():
             console.print(
                 "    [yellow]⚠[/] Skipped — GPU dependencies not installed.\n"
-                "    Install them with: [bold]pip install 'remove-ai-watermarks[gpu]'[/]"
+                "    Install them with: [bold]pip install 'remove-ai-watermarks\\[gpu]'[/]"
             )
         else:
             from remove_ai_watermarks.invisible_engine import InvisibleEngine
@@ -770,6 +793,7 @@ def cmd_all(
                 num_inference_steps=steps,
                 seed=seed,
                 humanize=humanize,
+                protect_text=not no_protect_text,
                 max_resolution=max_resolution,
             )
             console.print("    [green]✓[/] Invisible watermark removed")

src/remove_ai_watermarks/doubao_engine.py

@@ -239,7 +239,9 @@ class DoubaoEngine:
 
 def load_image_bgr(path: str | Path) -> NDArray:
     """Read an image as BGR ndarray (helper for scripts/tests)."""
-    img = cv2.imread(str(path), cv2.IMREAD_COLOR)
+    from remove_ai_watermarks import image_io
+
+    img = image_io.imread(path, cv2.IMREAD_COLOR)
     if img is None:
         raise FileNotFoundError(f"Failed to read image: {path}")
     return img

src/remove_ai_watermarks/gemini_engine.py

@@ -556,7 +556,9 @@ def detect_sparkle_confidence(image_path: Path) -> float | None:
     (cv2 returns None for unsupported containers such as HEIC). Kept here so the
     cv2 dependency stays in this module; callers apply their own threshold.
     """
-    img = cv2.imread(str(image_path))
+    from remove_ai_watermarks import image_io
+
+    img = image_io.imread(image_path)
     if img is None:
         return None
     return float(GeminiEngine().detect_watermark(img).confidence)

src/remove_ai_watermarks/identify.py

@@ -32,6 +32,7 @@ from remove_ai_watermarks.metadata import (
     exif_generator,
     get_ai_metadata,
     iptc_ai_system,
+    scan_head,
     xai_signature,
 )
 from remove_ai_watermarks.noai.c2pa import cbor_text_after, extract_c2pa_info, soft_binding_vendors_in
@@ -110,6 +111,11 @@ class ProvenanceReport:
     watermarks: list[str] = field(default_factory=list[str])
     signals: list[Signal] = field(default_factory=list["Signal"])
     caveats: list[str] = field(default_factory=list[str])
+    # Contradictions between independent provenance signals (e.g. two different
+    # AI vendors both claiming the image, or camera-capture credentials next to
+    # AI-generation markers). Non-empty means the provenance is internally
+    # inconsistent -- a strong tell of spoofed, transplanted, or laundered metadata.
+    integrity_clashes: list[str] = field(default_factory=list[str])
 
 
 def _issuers_in(data: bytes) -> list[str]:
@@ -188,6 +194,88 @@ def _attribute_platform(issuers: list[str], *, is_ai: bool = True) -> str | None
     return None
 
 
+# Coarse origin-vendor normalization for integrity-clash detection. Two signals
+# that resolve to the SAME key are consistent (a C2PA "Google (Gemini)" issuer
+# and a SynthID-Google proxy, or Adobe Firefly + its Adobe TrustMark soft
+# binding); two DIFFERENT keys from independent generator stamps are a
+# contradiction (a C2PA OpenAI manifest on an image whose EXIF says "Ideogram
+# AI"). Substring match on the lowercased platform/detail string; first hit wins,
+# so order specific tokens before brand umbrellas where they overlap.
+_AI_VENDOR_TOKENS: tuple[tuple[str, str], ...] = (
+    ("gpt-image", "OpenAI"),
+    ("dall", "OpenAI"),
+    ("sora", "OpenAI"),
+    ("openai", "OpenAI"),
+    ("gemini", "Google"),
+    ("imagen", "Google"),
+    ("nano banana", "Google"),
+    ("google", "Google"),
+    ("firefly", "Adobe"),
+    ("adobe", "Adobe"),
+    ("bing", "Microsoft"),
+    ("designer", "Microsoft"),
+    ("microsoft", "Microsoft"),
+    ("stability", "Stability AI"),
+    ("stable diffusion", "Stability AI"),
+    ("sdxl", "Stability AI"),
+    ("ideogram", "Ideogram"),
+    ("grok", "xAI"),
+    ("aurora", "xAI"),
+    ("xai", "xAI"),
+)
+
+
+def _vendor_of(text: str | None) -> str | None:
+    """Normalize a platform/generator string to a coarse origin-vendor key, or None."""
+    if not text:
+        return None
+    low = text.lower()
+    for token, vendor in _AI_VENDOR_TOKENS:
+        if token in low:
+            return vendor
+    return None
+
+
+def _integrity_clashes(
+    ai_vendors: dict[str, str], camera_label: str | None, *, camera_has_ai_marker: bool
+) -> list[str]:
+    """Surface contradictions between independent provenance signals.
+
+    Args:
+        ai_vendors: family name -> normalized AI-origin vendor, one entry per
+            generator-stamped signal (C2PA issuer when the source is AI, SynthID
+            proxy, EXIF/XMP generator tag, IPTC AISystemUsed, xAI, AIGC label).
+        camera_label: a camera/verified-capture C2PA device platform, if one was
+            identified (Pixel, Leica, Sony, Nikon, Truepic), else None.
+        camera_has_ai_marker: True when an AI-generation stamp coexists with the
+            camera credentials.
+
+    Returns:
+        Human-readable clash descriptions; empty when the signals agree.
+    """
+    clashes: list[str] = []
+
+    by_vendor: dict[str, list[str]] = {}
+    for family, vendor in ai_vendors.items():
+        by_vendor.setdefault(vendor, []).append(family)
+    if len(by_vendor) >= 2:
+        parts = [f"{vendor} (via {', '.join(sorted(fams))})" for vendor, fams in sorted(by_vendor.items())]
+        clashes.append(
+            "Conflicting AI-origin attributions from independent signals: "
+            + " vs ".join(parts)
+            + " -- one provenance set was likely spoofed, transplanted, or laundered."
+        )
+
+    if camera_label and camera_has_ai_marker:
+        vendors = ", ".join(sorted(set(ai_vendors.values()))) or "present"
+        clashes.append(
+            f"Camera-capture C2PA credentials ({camera_label}) coexist with AI-generation markers "
+            f"({vendors}) -- a genuine camera capture is not AI-generated, so the provenance is inconsistent."
+        )
+
+    return clashes
+
+
 def _visible_sparkle(image_path: Path) -> float | None:
     """Visible Gemini-sparkle confidence in [0, 1], or None if unavailable.
 
@@ -245,12 +333,20 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
 
     # First MB covers C2PA (PNG caBX, JPEG APP11, AVIF/HEIF/JXL uuid box) and
     # IPTC markers for the non-PNG path where extract_c2pa_info returns {}.
-    with open(image_path, "rb") as f:
-        head = f.read(_SCAN_BYTES)
+    # scan_head also seeks out late ISOBMFF provenance boxes (manifest after a
+    # large mdat in a streaming MP4) that a fixed first-MB read would miss.
+    head = scan_head(image_path, _SCAN_BYTES)
 
     signals: list[Signal] = []
     watermarks: list[str] = []
     caveats: list[str] = []
+    # One normalized origin vendor per generator-stamped signal, for integrity-
+    # clash detection (see _integrity_clashes). Visible sparkle and the open
+    # invisible watermark are deliberately excluded: the former is a fuzzy visual
+    # score, the latter can be a by-product of our own SDXL removal pass, so
+    # neither is a trustworthy "the generator stamped its identity" claim.
+    ai_vendor_claims: dict[str, str] = {}
+    camera_label = _device_platform(head)
 
     # ── C2PA Content Credentials ────────────────────────────────────
     has_c2pa = bool(info) or b"c2pa" in head.lower() or C2PA_UUID in head
@@ -271,11 +367,17 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
     # signer/producer), with the issuer byte-scan only as fallback. The issuer
     # scan alone mis-attributed real samples (Leica->Truepic timestamp authority,
     # Nikon->Adobe namespace, Pixel->Google Gemini) -- the device scan fixes that.
-    platform = (_device_platform(head) or _attribute_platform(issuers, is_ai=c2pa_is_ai)) if has_c2pa else None
+    platform = (camera_label or _attribute_platform(issuers, is_ai=c2pa_is_ai)) if has_c2pa else None
     if has_c2pa:
         detail = ", ".join(filter(None, [", ".join(issuers), generator, info.get("source_type")]))
         signals.append(Signal("c2pa", detail or "C2PA manifest present", "high"))
         watermarks.append(f"C2PA Content Credentials ({', '.join(issuers) or 'unknown signer'})")
+        # Record the AI-origin vendor for clash detection only when the source is
+        # actually AI -- classify the issuer attribution / generator, NOT the
+        # resolved `platform` (which may be a camera device token whose label,
+        # e.g. "Google Pixel", would mis-normalize to an AI vendor).
+        if c2pa_is_ai and (v := (_vendor_of(_attribute_platform(issuers, is_ai=True)) or _vendor_of(generator))):
+            ai_vendor_claims["c2pa"] = v
 
     # ── SynthID metadata proxy ──────────────────────────────────────
     # get_ai_metadata already sets synthid_watermark for both PNG (caBX parser)
@@ -286,6 +388,8 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
         caveats.append(_SYNTHID_CAVEAT)
         if "OpenAI" in (" ".join(issuers) + synthid):
             caveats.append(_OPENAI_CAVEAT)
+        if v := _vendor_of(synthid):
+            ai_vendor_claims["synthid"] = v
 
     # ── C2PA soft-binding: a named forensic/third-party watermark vendor ─
     # (Adobe TrustMark, Digimarc, Imatag, ...). Present in the manifest even when
@@ -315,6 +419,8 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
         watermarks.append(f"IPTC 2025.1 AI disclosure ({system})" if named else "IPTC 2025.1 AI disclosure fields")
         if platform is None and named:
             platform = f"{system} (IPTC AISystemUsed)"
+        if named and (v := _vendor_of(system)):
+            ai_vendor_claims["iptc_ai_system"] = v
 
     # ── China TC260 AIGC label (Doubao and other China-served gens) ──
     aigc = any(m in head for m in AIGC_MARKERS)
@@ -324,6 +430,7 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
         watermarks.append("China AIGC label (TC260 standard)")
         if platform is None:
             platform = "China AIGC-labeled generator (TC260; e.g. Doubao)"
+        ai_vendor_claims["aigc"] = "China AIGC (TC260)"
 
     # ── Local diffusion parameters (Stable Diffusion / ComfyUI) ──────
     local_keys = sorted(k for k in meta if k.lower() in _LOCAL_GEN_KEYS)
@@ -340,6 +447,8 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
         watermarks.append(f"Embedded generator tag: {generator_tag}")
         if platform is None:
             platform = f"{generator_tag} (EXIF/XMP generator tag)"
+        if v := _vendor_of(generator_tag):
+            ai_vendor_claims["exif_generator"] = v
 
     # ── xAI / Grok EXIF signature scheme (no C2PA/SynthID/IPTC) ──────
     # Grok's only provenance signal: EXIF ImageDescription "Signature: <base64>"
@@ -350,6 +459,7 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
         watermarks.append("xAI/Grok EXIF signature")
         if platform is None:
             platform = "xAI (Grok / Aurora)"
+        ai_vendor_claims["xai"] = "xAI"
 
     # ── Open invisible watermark (SD / SDXL / FLUX, dwtDct) ──────────
     # Public decoder, no key -- a definitive embedded signal on pristine files.
@@ -404,6 +514,9 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
         is_ai = None
         confidence = "none"
 
+    # ── Integrity clashes: contradictions between independent signals ─
+    clashes = _integrity_clashes(ai_vendor_claims, camera_label, camera_has_ai_marker=bool(ai_vendor_claims))
+
     caveats.append(_STRIP_CAVEAT)
     # De-duplicate while preserving order.
     caveats = list(dict.fromkeys(caveats))
@@ -416,4 +529,5 @@ def identify(image_path: Path, *, check_visible: bool = True, check_invisible: b
         watermarks=watermarks,
         signals=signals,
         caveats=caveats,
+        integrity_clashes=clashes,
     )

src/remove_ai_watermarks/image_io.py

@@ -0,0 +1,66 @@
+"""Unicode-safe cv2 image IO (issue #17).
+
+``cv2.imread`` / ``cv2.imwrite`` pass the path to the platform C runtime, which
+on Windows uses the narrow (ANSI) code-page API and therefore fails on paths
+containing non-ASCII characters (Chinese, Cyrillic, ...). The symptom is a
+``can't open/read file`` warning and a ``None`` decode even though the file
+exists.
+
+These wrappers route through numpy buffers instead: ``np.fromfile`` /
+``ndarray.tofile`` open the path in Python (full Unicode), and
+``cv2.imdecode`` / ``cv2.imencode`` do the codec work. The decoded/encoded
+bytes are byte-for-byte identical to ``imread`` / ``imwrite``. On macOS/Linux
+cv2 already accepts UTF-8 paths, so the wrappers are behavior-neutral there.
+
+cv2/numpy are imported lazily inside the functions so importing this module
+stays cheap in a bare environment (matching the rest of the package).
+"""
+
+# cv2 ships no type stubs; mirror the pragma used by the other cv2-using modules.
+# pyright: reportMissingTypeStubs=false, reportUnknownMemberType=false, reportUnknownVariableType=false, reportUnknownArgumentType=false
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import TYPE_CHECKING, Any
+
+if TYPE_CHECKING:
+    from numpy.typing import NDArray
+
+
+def imread(path: str | Path, flags: int | None = None) -> NDArray[Any] | None:
+    """Unicode-safe ``cv2.imread``.
+
+    ``flags`` defaults to ``cv2.IMREAD_COLOR`` (same as ``cv2.imread``). Returns
+    ``None`` when the file is missing or cannot be decoded, matching
+    ``cv2.imread`` semantics so existing ``if img is None`` checks keep working.
+    """
+    import cv2
+    import numpy as np
+
+    if flags is None:
+        flags = cv2.IMREAD_COLOR
+    try:
+        data = np.fromfile(str(path), dtype=np.uint8)
+    except OSError:
+        return None
+    if data.size == 0:
+        return None
+    return cv2.imdecode(data, flags)
+
+
+def imwrite(path: str | Path, img: NDArray[Any]) -> bool:
+    """Unicode-safe ``cv2.imwrite``.
+
+    The output format is taken from the path extension (e.g. ``.png``), exactly
+    like ``cv2.imwrite``. Returns ``True`` on success, ``False`` if the codec
+    rejects the image.
+    """
+    import cv2
+
+    ext = Path(path).suffix or ".png"
+    ok, buf = cv2.imencode(ext, img)
+    if not ok:
+        return False
+    buf.tofile(str(path))
+    return True

src/remove_ai_watermarks/invisible_engine.py

@@ -125,6 +125,7 @@ class InvisibleEngine:
         seed: int | None = None,
         humanize: float = 0.0,
         protect_faces: bool = True,
+        protect_text: bool = True,
         max_resolution: int = 0,
     ) -> Path:
         """Remove invisible watermark from an image.
@@ -138,6 +139,9 @@ class InvisibleEngine:
             seed: Random seed for reproducibility.
             humanize: Intensity of Analog Humanizer film grain (0 = off).
             protect_faces: Boolean to extract and restore faces intact.
+            protect_text: Detect text regions and preserve them via Differential
+                Diffusion when any are found, so glyphs (incl. CJK) survive the
+                removal pass. On by default; the detector decides per image.
             max_resolution: Cap the long side (px) before diffusion. 0 (default)
                 = native resolution, no pre-downscale -- matches the hosted
                 raiw.cc backend. Set a positive value only to bound GPU/MPS
@@ -210,6 +214,7 @@ class InvisibleEngine:
                 num_inference_steps=num_inference_steps,
                 guidance_scale=guidance_scale,
                 seed=seed,
+                protect_text=protect_text,
             )
 
             # Optional: Face restoration & Humanizer (Phase 2 - Post-processing)
@@ -217,7 +222,9 @@ class InvisibleEngine:
                 import cv2
                 import numpy as np
 
-                out_cv = cv2.imread(str(out_path), cv2.IMREAD_COLOR)
+                from remove_ai_watermarks import image_io
+
+                out_cv = image_io.imread(out_path, cv2.IMREAD_COLOR)
 
                 if protect_faces and original_faces:
                     if self._progress_callback:
@@ -243,20 +250,22 @@ class InvisibleEngine:
                     # Using INTER_LANCZOS4 for high-quality upscaling back to original
                     out_cv = cv2.resize(out_cv, orig_size, interpolation=cv2.INTER_LANCZOS4)
 
-                cv2.imwrite(str(out_path), out_cv)
+                image_io.imwrite(out_path, out_cv)
 
             else:
                 # Even if no protect_faces or humanize, we must restore original size if needed
                 import cv2
 
-                out_cv = cv2.imread(str(out_path), cv2.IMREAD_COLOR)
+                from remove_ai_watermarks import image_io
+
+                out_cv = image_io.imread(out_path, cv2.IMREAD_COLOR)
                 if out_cv is not None and (out_cv.shape[1], out_cv.shape[0]) != orig_size:
                     if self._progress_callback:
                         self._progress_callback(
                             f"Upscaling result back to original resolution {orig_size[0]}x{orig_size[1]}..."
                         )
                     out_cv = cv2.resize(out_cv, orig_size, interpolation=cv2.INTER_LANCZOS4)
-                    cv2.imwrite(str(out_path), out_cv)
+                    image_io.imwrite(out_path, out_cv)
 
             return out_path
         finally:

src/remove_ai_watermarks/invisible_watermark.py

@@ -78,10 +78,11 @@ def detect_invisible_watermark(image_path: Path) -> str | None:
     """
     if not is_available():
         return None
-    import cv2
     from imwatermark import WatermarkDecoder
 
-    img = cv2.imread(str(image_path))
+    from remove_ai_watermarks import image_io
+
+    img = image_io.imread(image_path)
     if img is None:
         return None
 

src/remove_ai_watermarks/metadata.py

@@ -132,6 +132,28 @@ def _is_ai_key(key: str) -> bool:
     return any(kw in key_lower for kw in AI_KEYWORDS)
 
 
+def scan_head(image_path: Path, size: int = 1024 * 1024) -> bytes:
+    """First ``size`` bytes of the file, plus -- for ISOBMFF containers -- the
+    payloads of any provenance (``uuid`` / ``jumb``) boxes found beyond that
+    window by seeking past large boxes like ``mdat``.
+
+    This is the shared input for every C2PA / AIGC / IPTC byte scan. The
+    ISOBMFF extension catches a manifest placed AFTER the media data in a
+    streaming / non-faststart MP4, which a fixed first-MB read would miss. For
+    non-ISOBMFF inputs it is exactly ``f.read(size)`` -- behavior-neutral.
+    """
+    with open(image_path, "rb") as f:
+        head = f.read(size)
+    # Lazy import: isobmff imports this module's constants at top level.
+    from remove_ai_watermarks.noai import isobmff
+
+    if isobmff.is_isobmff(head):
+        region = isobmff.scan_c2pa_region(image_path)
+        if region:
+            head += region
+    return head
+
+
 def has_ai_metadata(image_path: Path) -> bool:
     """Check if an image contains AI-generation metadata.
 
@@ -167,9 +189,8 @@ def has_ai_metadata(image_path: Path) -> bool:
         pass
 
     # Binary scan covers C2PA (PNG caBX, JPEG APP11, AVIF/HEIF/JXL uuid boxes)
-    # and IPTC AI markers in XMP. Read only the first 512KB to bound memory.
-    with open(image_path, "rb") as f:
-        data = f.read(512 * 1024)
+    # and IPTC AI markers in XMP. First 512KB (plus late ISOBMFF provenance boxes).
+    data = scan_head(image_path, 512 * 1024)
     if b"c2pa" in data.lower() or b"C2PA" in data:
         return True
     if C2PA_UUID in data:
@@ -196,8 +217,7 @@ def aigc_label(image_path: Path) -> dict[str, str] | None:
     import json
     import re
 
-    with open(image_path, "rb") as f:
-        data = f.read(1024 * 1024)
+    data = scan_head(image_path)
     match = re.search(rb"<TC260:AIGC>(.*?)</TC260:AIGC>", data, re.DOTALL)
     if not match:
         return None
@@ -219,8 +239,7 @@ def iptc_ai_system(image_path: Path) -> str | None:
     extractable, otherwise the literal ``"fields present"``. Container-agnostic
     raw-byte scan; handles both XMP element and attribute serializations.
     """
-    with open(image_path, "rb") as f:
-        data = f.read(1024 * 1024)
+    data = scan_head(image_path)
     if not any(marker in data for marker in IPTC_AI_FIELD_MARKERS):
         return None
     match = re.search(rb"AISystemUsed[=:\s]*[\"'>]\s*([^<\"']{1,120})", data)
@@ -259,8 +278,7 @@ def synthid_source(image_path: Path) -> str | None:
     # Non-PNG containers (JPEG APP11, WebP, AVIF/HEIF/JXL uuid box) keep the
     # C2PA manifest where the PNG parser can't reach it. Binary-scan for the
     # same signal: a C2PA manifest from a SynthID-using issuer on AI content.
-    with open(image_path, "rb") as f:
-        data = f.read(1024 * 1024)
+    data = scan_head(image_path)
     has_c2pa = b"c2pa" in data.lower() or C2PA_UUID in data
     # Matches both "trainedAlgorithmicMedia" and "compositeWithTrainedAlgorithmicMedia".
     ai_source = b"trainedAlgorithmicMedia" in data or b"TrainedAlgorithmicMedia" in data
@@ -311,8 +329,7 @@ def exif_generator(image_path: Path) -> str | None:
 
     # XMP CreatorTool: text, container-agnostic (covers HEIF/JXL via raw scan).
     try:
-        with open(image_path, "rb") as f:
-            head = f.read(1024 * 1024)
+        head = scan_head(image_path)
         for match in re.finditer(rb"CreatorTool[>\"'=\s]{1,4}([^<\"']{1,80})", head):
             candidates.append(match.group(1).decode("latin1", "replace"))
     except Exception as exc:
@@ -467,8 +484,7 @@ def get_ai_metadata(image_path: Path) -> dict[str, str]:
     if "synthid_watermark" not in result and (vendor := synthid_source(image_path)):
         result.setdefault("synthid_watermark", synthid_verdict(vendor))
     if "soft_binding" not in result:
-        with open(image_path, "rb") as f:
-            head = f.read(1024 * 1024)
+        head = scan_head(image_path)
         if vendors := soft_binding_vendors_in(head):
             result["soft_binding"] = ", ".join(vendors)
 
@@ -507,10 +523,18 @@ def _strip_with_ffmpeg(source_path: Path, output_path: Path) -> Path:
         )
     output_path.parent.mkdir(parents=True, exist_ok=True)
     cmd = [
-        ffmpeg, "-y", "-loglevel", "error",
-        "-i", str(source_path),
-        "-map_metadata", "-1", "-map_chapters", "-1",
-        "-c", "copy",
+        ffmpeg,
+        "-y",
+        "-loglevel",
+        "error",
+        "-i",
+        str(source_path),
+        "-map_metadata",
+        "-1",
+        "-map_chapters",
+        "-1",
+        "-c",
+        "copy",
         str(output_path),
     ]
     result = subprocess.run(cmd, capture_output=True, text=True, check=False)  # noqa: S603
@@ -551,16 +575,25 @@ def remove_ai_metadata(
     # codestream bit-for-bit. MP4/MOV/M4A are ISOBMFF too, so the same top-level
     # uuid/jumb box walker applies. Route by suffix OR by an ``ftyp`` content
     # sniff, so a correctly-shaped container is handled whatever its extension.
-    from remove_ai_watermarks.noai.isobmff import is_isobmff, strip_c2pa_boxes
+    from remove_ai_watermarks.noai.isobmff import blank_ai_xmp_packets, is_isobmff, strip_c2pa_boxes
 
     with open(source_path, "rb") as f:
         head = f.read(12)
     if source_path.suffix.lower() in _ISOBMFF_EXTS or is_isobmff(head):
         data = source_path.read_bytes()
+        # Top-level uuid/jumb boxes (C2PA + AI-label XMP), then AI-label XMP that
+        # lives inside a meta-box ``mime`` item (HEIF/AVIF) -- blanked in place so
+        # box sizes and iloc offsets stay valid and the coded image is untouched.
         cleaned, stripped = strip_c2pa_boxes(data)
+        cleaned, blanked = blank_ai_xmp_packets(cleaned)
         output_path.parent.mkdir(parents=True, exist_ok=True)
         output_path.write_bytes(cleaned)
-        logger.info("Stripped %d AI-provenance box(es) → %s", stripped, output_path)
+        logger.info(
+            "Stripped %d AI-provenance box(es), blanked %d meta-box XMP packet(s) → %s",
+            stripped,
+            blanked,
+            output_path,
+        )
         return output_path
 
     # Non-ISOBMFF audio/video (WebM/Matroska EBML, MP3 ID3, WAV/FLAC/OGG): the

src/remove_ai_watermarks/noai/img2img_runner.py

@@ -121,6 +121,96 @@ def run_img2img_with_mps_fallback(
         raise
 
 
+def run_differential(
+    pipeline: Any,
+    image: Image.Image,
+    change_map: Any,
+    strength: float,
+    num_inference_steps: int,
+    guidance_scale: float,
+    generator: Any,
+    device: str,
+    set_progress: Callable[[str], None],
+) -> Image.Image:
+    """Run the SDXL Differential-Diffusion pipeline and return the image.
+
+    Unlike standard img2img, the differential pipeline needs pre-processed image
+    tensors plus a per-pixel change map (HxW float32 in [0, 1]); white preserves
+    the original pixels, black regenerates them. Runs without a step callback --
+    the community pipeline's callback signature differs across diffusers
+    versions, and a protect-text pass is short.
+    """
+    import torch
+
+    image_tensor = pipeline.image_processor.preprocess(image).to(device)
+    map_tensor = torch.from_numpy(change_map)[None].to(device)  # pyright: ignore[reportPrivateImportUsage, reportUnknownMemberType]
+    set_progress(f"Running protected regeneration ({device}, strength={strength})...")
+    result = pipeline(
+        prompt="",
+        image=image_tensor,
+        original_image=image_tensor,
+        map=map_tensor,
+        strength=strength,
+        num_inference_steps=num_inference_steps,
+        guidance_scale=guidance_scale,
+        generator=generator,
+    )
+    return result.images[0]
+
+
+def run_differential_with_mps_fallback(
+    load_pipeline: Callable[[], Any],
+    image: Image.Image,
+    change_map: Any,
+    strength: float,
+    num_inference_steps: int,
+    guidance_scale: float,
+    generator: Any,
+    device: str,
+    set_progress: Callable[[str], None],
+    *,
+    reload_on_cpu: Callable[[], Any],
+) -> tuple[Image.Image, str]:
+    """Run differential img2img; on MPS error, fall back to CPU.
+
+    Returns:
+        (result_image, final_device) -- device may change to ``"cpu"`` on fallback.
+    """
+    pipeline = load_pipeline()
+    try:
+        img = run_differential(
+            pipeline,
+            image,
+            change_map,
+            strength,
+            num_inference_steps,
+            guidance_scale,
+            generator,
+            device,
+            set_progress,
+        )
+        return img, device
+    except RuntimeError as error:
+        if device == "mps" and is_mps_error(error):
+            logger.warning("MPS error detected: %s. Falling back to CPU.", error)
+            set_progress("MPS error! Clearing cache and retrying on CPU...")
+            _try_clear_mps_cache()
+            pipeline = reload_on_cpu()
+            img = run_differential(
+                pipeline,
+                image,
+                change_map,
+                strength,
+                num_inference_steps,
+                guidance_scale,
+                None,
+                "cpu",
+                set_progress,
+            )
+            return img, "cpu"
+        raise
+
+
 def _call_pipeline(
     pipeline: Any,
     image: Image.Image,

src/remove_ai_watermarks/noai/isobmff.py

@@ -17,11 +17,13 @@ Reference: ISO/IEC 14496-12 (ISOBMFF) and C2PA 2.1 spec §11.
 
 from __future__ import annotations
 
+import re
 import struct
 from typing import TYPE_CHECKING
 
 if TYPE_CHECKING:
     from collections.abc import Iterator
+    from pathlib import Path
 
 from remove_ai_watermarks.metadata import (
     AIGC_MARKERS,
@@ -42,6 +44,12 @@ C2PA_BOX_TYPES: frozenset[bytes] = frozenset({b"uuid", b"jumb"})
 # XMP (copyright, camera info) is kept.
 _AI_LABEL_MARKERS: tuple[bytes, ...] = AIGC_MARKERS + IPTC_AI_MARKERS + IPTC_AI_FIELD_MARKERS
 
+# Adobe XMP packet delimiters (XMP spec part 3). In HEIF/AVIF the XMP packet
+# sits inside a ``meta``-box ``mime`` item whose bytes live in ``mdat`` / ``idat``,
+# out of reach of the top-level box stripper, so an AI-label packet there is
+# blanked in place (see ``blank_ai_xmp_packets``).
+_XMP_PACKET_RE = re.compile(rb"<\?xpacket begin=.*?<\?xpacket end=[^>]*?\?>", re.DOTALL)
+
 
 def _iter_top_level_boxes(data: bytes) -> Iterator[tuple[int, int, bytes, int]]:
     """Yield ``(start, end, type, payload_offset)`` for each top-level box.
@@ -78,6 +86,58 @@ def is_isobmff(data: bytes) -> bool:
     return len(data) >= 8 and data[4:8] == b"ftyp"
 
 
+def scan_c2pa_region(path: str | Path, *, max_total: int = 4 * 1024 * 1024) -> bytes:
+    """Concatenated payloads of top-level ``uuid`` / ``jumb`` boxes in an ISOBMFF
+    file, found by seeking past other boxes (``mdat`` etc.) by size.
+
+    C2PA manifests and XMP packets (incl. AI labels) live in top-level ``uuid``
+    boxes; JPEG-XL uses ``jumb``. In a streaming / non-faststart MP4 the manifest
+    sits AFTER a multi-megabyte ``mdat``, so a fixed first-MB read misses it. This
+    walks box headers (8-16 bytes each) and seeks past payloads it does not need,
+    so it never loads ``mdat`` into memory and works on multi-GB files. Returns
+    the relevant box payloads (capped at ``max_total``), or ``b""`` for a
+    non-ISOBMFF file or on any read error.
+    """
+    collected = bytearray()
+    try:
+        with open(path, "rb") as f:
+            sniff = f.read(8)
+            if len(sniff) < 8 or sniff[4:8] != b"ftyp":
+                return b""
+            f.seek(0, 2)
+            file_size = f.tell()
+            pos = 0
+            while pos + 8 <= file_size and len(collected) < max_total:
+                f.seek(pos)
+                header = f.read(8)
+                if len(header) < 8:
+                    break
+                size32 = struct.unpack(">I", header[:4])[0]
+                box_type = header[4:8]
+                payload_off = pos + 8
+                if size32 == 1:
+                    ext = f.read(8)
+                    if len(ext) < 8:
+                        break
+                    size = struct.unpack(">Q", ext)[0]
+                    payload_off = pos + 16
+                elif size32 == 0:
+                    size = file_size - pos
+                else:
+                    size = size32
+                if size < (payload_off - pos) or pos + size > file_size:
+                    break
+                if box_type in C2PA_BOX_TYPES:
+                    f.seek(payload_off)
+                    to_read = min(pos + size - payload_off, max_total - len(collected))
+                    if to_read > 0:
+                        collected += f.read(to_read)
+                pos += size
+    except OSError:
+        return b""
+    return bytes(collected)
+
+
 def strip_c2pa_boxes(data: bytes) -> tuple[bytes, int]:
     """Return ``(cleaned_bytes, stripped_count)`` with AI-provenance boxes removed.
 
@@ -92,9 +152,10 @@ def strip_c2pa_boxes(data: bytes) -> tuple[bytes, int]:
     All other boxes (incl. ``mdat`` / codestream) are emitted verbatim, so pixel
     and audio data is preserved bit-for-bit. Non-ISOBMFF input is returned
     unchanged. Despite the name this also covers MP4/MOV/M4A video and audio
-    (all ISOBMFF). NOTE: EXIF/XMP stored as *items inside the ``meta`` box*
-    (typical for AVIF/HEIF images) is not removed -- that needs meta-box surgery
-    and is a documented limitation.
+    (all ISOBMFF). NOTE: this drops only top-level boxes. An AI-label XMP packet
+    stored as an *item inside the ``meta`` box* (typical for AVIF/HEIF) is handled
+    separately by :func:`blank_ai_xmp_packets`; an ``Exif`` meta-box item is still
+    not removed (would need meta-box surgery) and remains a documented limitation.
     """
     if not is_isobmff(data):
         return data, 0
@@ -114,3 +175,30 @@ def strip_c2pa_boxes(data: bytes) -> tuple[bytes, int]:
             continue
         out.extend(data[start:end])
     return bytes(out), stripped
+
+
+def blank_ai_xmp_packets(data: bytes) -> tuple[bytes, int]:
+    """Overwrite (with spaces, in place) any XMP packet carrying an AI-label
+    marker; return ``(data, blanked_count)``.
+
+    HEIF/AVIF store XMP as a ``meta``-box ``mime`` item whose bytes live in
+    ``mdat`` / ``idat``, which ``strip_c2pa_boxes`` cannot remove without
+    meta-box surgery (``iinf`` / ``iloc`` rewrite). Instead, the XMP packet is
+    located by its ``<?xpacket begin ... end?>`` delimiters and, when it carries
+    an AI-label marker (TC260 AIGC / IPTC / IPTC-2025.1), overwritten with spaces.
+    Because the replacement is the **same length**, every box size and ``iloc``
+    offset stays valid and the coded image data is untouched -- only the AI label
+    content is destroyed. Packets without an AI marker (plain copyright / camera
+    XMP) are left intact, mirroring the top-level XMP-``uuid`` content match.
+    """
+    blanked = 0
+
+    def _scrub(match: re.Match[bytes]) -> bytes:
+        nonlocal blanked
+        packet = match.group()
+        if any(marker in packet for marker in _AI_LABEL_MARKERS):
+            blanked += 1
+            return b" " * len(packet)
+        return packet
+
+    return _XMP_PACKET_RE.sub(_scrub, data), blanked

src/remove_ai_watermarks/noai/watermark_remover.py

@@ -229,6 +229,13 @@ def get_device() -> str:
 # Keep legacy name available for backwards compatibility
 _detect_model_profile_from_id = detect_model_profile
 
+# SDXL Differential-Diffusion community pipeline, pinned to the installed
+# diffusers version so the fetched pipeline code matches the library (see #21).
+# Diffusers' dynamic-module loader resolves ``custom_revision`` against the
+# package version string (``0.38.0``), NOT the GitHub git tag (``v0.38.0``).
+_DIFF_PIPELINE_NAME = "pipeline_stable_diffusion_xl_differential_img2img"
+_DIFF_PIPELINE_REVISION = "0.38.0"
+
 
 class WatermarkRemover:
     """Remove watermarks from images using diffusion model regeneration.
@@ -271,6 +278,7 @@ class WatermarkRemover:
             self.torch_dtype = torch_dtype
 
         self._pipeline: AutoImg2ImgPipeline | None = None
+        self._diff_pipeline: Any = None
         self._ctrlregen_engine: Any = None
         self._progress_callback = progress_callback
         self.hf_token: str | None = hf_token or os.environ.get("HF_TOKEN")
@@ -379,6 +387,7 @@ class WatermarkRemover:
         num_inference_steps: int = 50,
         guidance_scale: float | None = None,
         seed: int | None = None,
+        protect_text: bool = True,
     ) -> Path:
         """Remove watermark from an image using regeneration attack.
 
@@ -389,6 +398,10 @@ class WatermarkRemover:
             num_inference_steps: Number of denoising steps.
             guidance_scale: Classifier-free guidance scale.
             seed: Random seed for reproducibility.
+            protect_text: Detect text regions and preserve them via Differential
+                Diffusion when any are found (SDXL default profile only). On by
+                default; the detector decides per image, and text-free inputs run
+                the standard pass at no extra cost.
 
         Returns:
             Path to the cleaned image.
@@ -437,7 +450,21 @@ class WatermarkRemover:
                 guidance_scale,
                 generator,
             )
+        elif protect_text and self._can_protect_text():
+            cleaned_image = self._run_differential(
+                init_image,
+                strength,
+                num_inference_steps,
+                guidance_scale,
+                generator,
+            )
         else:
+            if protect_text:
+                logger.debug(
+                    "Text protection unavailable "
+                    "(needs the SDXL default model and the cv2 text detector); "
+                    "running standard img2img."
+                )
             cleaned_image = self._run_img2img(
                 init_image,
                 strength,
@@ -520,6 +547,99 @@ class WatermarkRemover:
         self._pipeline = None
         return self._load_pipeline()
 
+    # ── Text-protected differential runner ───────────────────────────
+
+    def _can_protect_text(self) -> bool:
+        """True when text protection can run: SDXL default model + cv2 detector."""
+        from remove_ai_watermarks import text_protector
+
+        return self.model_id == self.DEFAULT_MODEL_ID and text_protector.is_available()
+
+    def _load_differential_pipeline(self) -> Any:
+        """Load the SDXL Differential-Diffusion community pipeline lazily."""
+        if self._diff_pipeline is None:
+            from diffusers import DiffusionPipeline
+
+            self._set_progress("Loading Differential-Diffusion pipeline (protect-text)...")
+            use_fp16 = self.device in {"mps", "cuda"}
+            load_kwargs: dict[str, Any] = {
+                "custom_pipeline": _DIFF_PIPELINE_NAME,
+                "custom_revision": _DIFF_PIPELINE_REVISION,
+                "torch_dtype": torch.float16 if use_fp16 else torch.float32,  # type: ignore[attr-defined]
+                "use_safetensors": True,
+            }
+            if use_fp16:
+                load_kwargs["variant"] = "fp16"
+            if self.hf_token:
+                load_kwargs["token"] = self.hf_token
+
+            pipeline = DiffusionPipeline.from_pretrained(self.model_id, **load_kwargs).to(self.device)
+            # The differential pipeline upcasts the SDXL VAE to fp32 internally
+            # (the fp16 VAE decodes to NaN/black otherwise), so we add no extra
+            # VAE handling here. Attention slicing is also left off on MPS: it
+            # produced NaN latents with this pipeline, and the protect-text pass
+            # is short enough not to need it.
+            with contextlib.suppress(Exception):
+                pipeline.set_progress_bar_config(disable=True)
+            self._diff_pipeline = pipeline
+        return self._diff_pipeline
+
+    def _reload_differential_on_cpu(self) -> Any:
+        """Reload the differential pipeline on CPU after an MPS failure."""
+        self.device = "cpu"
+        self.torch_dtype = torch.float32  # type: ignore[assignment]
+        self._diff_pipeline = None
+        return self._load_differential_pipeline()
+
+    def _run_differential(
+        self,
+        init_image: Image.Image,
+        strength: float,
+        num_inference_steps: int,
+        guidance_scale: float,
+        generator: Any,
+    ) -> Image.Image:
+        """Run differential img2img that preserves detected text regions."""
+        import cv2
+        import numpy as np
+
+        from remove_ai_watermarks import text_protector
+
+        self._set_progress("Detecting text regions to protect (protect-text)...")
+        bgr = cv2.cvtColor(np.array(init_image), cv2.COLOR_RGB2BGR)
+        try:
+            boxes = text_protector.TextProtector().detect_text_boxes(bgr)
+        except Exception as exc:
+            logger.warning("Text detection failed (%s); running standard img2img.", exc)
+            return self._run_img2img(init_image, strength, num_inference_steps, guidance_scale, generator)
+
+        if not boxes:
+            self._set_progress("No text detected; running standard img2img.")
+            return self._run_img2img(init_image, strength, num_inference_steps, guidance_scale, generator)
+
+        width, height = init_image.size
+        change_map = text_protector.build_change_map(boxes, height, width)
+        self._set_progress(f"Protecting {len(boxes)} text region(s) via Differential Diffusion...")
+
+        from remove_ai_watermarks.noai.img2img_runner import run_differential_with_mps_fallback
+
+        result_image, final_device = run_differential_with_mps_fallback(
+            load_pipeline=self._load_differential_pipeline,
+            image=init_image,
+            change_map=change_map,
+            strength=strength,
+            num_inference_steps=num_inference_steps,
+            guidance_scale=guidance_scale,
+            generator=generator,
+            device=self.device,
+            set_progress=self._set_progress,
+            reload_on_cpu=self._reload_differential_on_cpu,
+        )
+        if final_device != self.device:
+            self.device = final_device
+            self.torch_dtype = torch.float32  # type: ignore[assignment]
+        return result_image
+
     # ── CtrlRegen runner ─────────────────────────────────────────────
 
     def _run_ctrlregen(

src/remove_ai_watermarks/text_protector.py

@@ -0,0 +1,167 @@
+"""Text-region protection for diffusion-based watermark removal.
+
+SDXL img2img (the ``invisible`` pipeline) regenerates every pixel, so small text
+and CJK glyphs get deformed at the strengths that defeat SynthID (issue #21).
+This module detects text regions and builds a per-pixel "change map" for
+Differential Diffusion: the background is regenerated normally while text
+regions are largely preserved, so glyphs survive the watermark-removal pass.
+
+Detection uses only OpenCV's DNN module (no torch): the PP-OCRv3 text detector
+is a ~2.4 MB ONNX model (Apache-2.0, from opencv_zoo) that is CJK-native and
+returns rotated quadrilaterals. The model is downloaded and cached on first use;
+it is never bundled in this repo.
+
+Change-map polarity (verified empirically against the differential pipeline):
+white (1.0) = PRESERVE the original pixels, black (0.0) = MAXIMUM change. So the
+map is black everywhere except the text polygons, which are painted toward
+white. ``preserve`` stays below a hard 1.0 freeze by default: SynthID is
+designed to survive cropping, so totally freezing text pixels would leave the
+watermark intact there. A high-but-partial preserve still scrubs lightly.
+"""
+
+# cv2 ships no type stubs; mirror the pragma used by the other cv2-using modules.
+# pyright: reportMissingTypeStubs=false, reportUnknownMemberType=false, reportUnknownVariableType=false, reportUnknownArgumentType=false, reportCallIssue=false, reportArgumentType=false, reportReturnType=false
+
+from __future__ import annotations
+
+import logging
+import os
+import tempfile
+import urllib.request
+from pathlib import Path
+from typing import TYPE_CHECKING, Any
+
+if TYPE_CHECKING:
+    from numpy.typing import NDArray
+
+logger = logging.getLogger(__name__)
+
+# PP-OCRv3 Chinese text detector (DB head), opencv_zoo, Apache-2.0.
+_MODEL_URL = (
+    "https://github.com/opencv/opencv_zoo/raw/main/models/text_detection_ppocr/text_detection_cn_ppocrv3_2023may.onnx"
+)
+_MODEL_FILENAME = "text_detection_cn_ppocrv3_2023may.onnx"
+
+# DB detector input: long side scaled to this, rounded to a multiple of 32.
+_DET_INPUT_LONG_SIDE = 736
+# ImageNet mean (x255) and 1/255 scale -- the normalization PP-OCRv3 expects.
+_DET_MEAN = (0.485 * 255, 0.456 * 255, 0.406 * 255)
+_DET_SCALE = 1 / 255.0
+
+
+def is_available() -> bool:
+    """True when OpenCV's DNN text-detection model is importable."""
+    try:
+        import cv2
+
+        return hasattr(cv2.dnn, "TextDetectionModel_DB")
+    except ImportError:
+        return False
+
+
+def _cache_dir() -> Path:
+    """Local cache directory for the detector model (created on demand)."""
+    cache = Path.home() / ".cache" / "remove-ai-watermarks"
+    cache.mkdir(parents=True, exist_ok=True)
+    return cache
+
+
+def _model_path() -> Path:
+    """Return the cached detector path, downloading it on first use."""
+    target = _cache_dir() / _MODEL_FILENAME
+    if target.exists() and target.stat().st_size > 0:
+        return target
+    logger.info("Downloading PP-OCRv3 text detector (~2.4 MB) to %s", target)
+    # Download to a temp file in the same dir, then atomically rename so a
+    # partial download never leaves a corrupt model cached.
+    fd, tmp_name = tempfile.mkstemp(dir=str(target.parent), suffix=".onnx.part")
+    tmp_path = Path(tmp_name)
+    try:
+        os.close(fd)
+        with urllib.request.urlopen(_MODEL_URL) as resp:  # noqa: S310 (trusted GitHub URL)
+            tmp_path.write_bytes(resp.read())
+        tmp_path.replace(target)
+    finally:
+        if tmp_path.exists():
+            tmp_path.unlink()
+    return target
+
+
+def build_change_map(
+    boxes: list[NDArray[Any]],
+    height: int,
+    width: int,
+    preserve: float = 0.9,
+    feather: int = 15,
+) -> NDArray[Any]:
+    """Build a Differential-Diffusion change map from text polygons.
+
+    Args:
+        boxes: Text-region polygons as arrays of (x, y) vertices.
+        height: Output map height in pixels.
+        width: Output map width in pixels.
+        preserve: Map value painted inside text polygons (0..1). White (1.0)
+            fully preserves the original pixels; the default 0.9 preserves
+            strongly while still letting a light scrub through.
+        feather: Gaussian-blur kernel size for soft polygon edges (forced odd).
+
+    Returns:
+        Float32 HxW array in [0, 1]: ~0 in the background (full change),
+        ``preserve`` inside text regions, blended at the edges.
+    """
+    import cv2
+    import numpy as np
+
+    change_map = np.zeros((height, width), np.float32)
+    if boxes:
+        polys = [np.asarray(b, np.int32) for b in boxes]
+        cv2.fillPoly(change_map, polys, float(preserve))
+    if feather > 0:
+        if feather % 2 == 0:
+            feather += 1
+        change_map = cv2.GaussianBlur(change_map, (feather, feather), 0)
+        # GaussianBlur can overshoot the painted value by a float epsilon; keep
+        # the contract that the map stays a valid [0, 1] change map.
+        np.clip(change_map, 0.0, 1.0, out=change_map)
+    return change_map
+
+
+class TextProtector:
+    """Detect text regions with PP-OCRv3 for diffusion change-map protection."""
+
+    def __init__(
+        self,
+        binary_threshold: float = 0.3,
+        polygon_threshold: float = 0.5,
+        max_candidates: int = 200,
+        unclip_ratio: float = 2.0,
+    ) -> None:
+        import cv2
+
+        self._detector = cv2.dnn.TextDetectionModel_DB(str(_model_path()))
+        self._detector.setBinaryThreshold(binary_threshold)
+        self._detector.setPolygonThreshold(polygon_threshold)
+        self._detector.setMaxCandidates(max_candidates)
+        self._detector.setUnclipRatio(unclip_ratio)
+
+    def detect_text_boxes(self, bgr_image: NDArray[Any]) -> list[NDArray[Any]]:
+        """Detect text regions, returning a list of rotated quad polygons.
+
+        Args:
+            bgr_image: Image as an HxWx3 BGR uint8 array (OpenCV convention).
+
+        Returns:
+            One array of four (x, y) vertices per detected text region.
+        """
+        height, width = bgr_image.shape[:2]
+        scale = _DET_INPUT_LONG_SIDE / max(height, width)
+        in_w = max((round(width * scale) // 32) * 32, 32)
+        in_h = max((round(height * scale) // 32) * 32, 32)
+        self._detector.setInputParams(
+            scale=_DET_SCALE,
+            size=(in_w, in_h),
+            mean=_DET_MEAN,
+            swapRB=True,
+        )
+        boxes, _confidences = self._detector.detect(bgr_image)
+        return list(boxes)

tests/test_cli.py

@@ -509,3 +509,21 @@ class TestBatchCommand:
         assert result.exit_code == 0
         expected_dir = tmp_path / "input_clean"
         assert expected_dir.exists()
+
+
+class TestGpuHintMarkup:
+    """The GPU-extra install hint must survive rich markup (the ``[gpu]`` token
+    is otherwise parsed as a style tag and silently dropped)."""
+
+    def test_invisible_install_hint_keeps_gpu_extra(self, runner, sample_png):
+        with patch("remove_ai_watermarks.invisible_engine.is_available", return_value=False):
+            result = runner.invoke(main, ["invisible", str(sample_png)])
+        assert result.exit_code != 0
+        assert "remove-ai-watermarks[gpu]" in result.output
+
+    def test_all_install_hint_keeps_gpu_extra(self, runner, sample_png):
+        # The `all` pipeline skips the invisible step with a warning that carries
+        # the same hint; it must keep the [gpu] extra too.
+        with patch("remove_ai_watermarks.invisible_engine.is_available", return_value=False):
+            result = runner.invoke(main, ["all", str(sample_png)])
+        assert "remove-ai-watermarks[gpu]" in result.output

tests/test_identify.py

@@ -17,7 +17,9 @@ from remove_ai_watermarks.identify import (
     ProvenanceReport,
     _ai_tools_in,
     _attribute_platform,
+    _integrity_clashes,
     _issuers_in,
+    _vendor_of,
     identify,
 )
 
@@ -447,3 +449,101 @@ class TestIdentifyAIGC:
         r = identify(self._aigc_png(tmp_path), check_visible=False)
         sig = next(s for s in r.signals if s.name == "aigc")
         assert "BYTEDANCE001" in sig.detail
+
+
+# ── Integrity clashes (contradictions between independent signals) ──────
+
+
+class TestVendorOf:
+    def test_openai_variants(self):
+        assert _vendor_of("OpenAI (ChatGPT / gpt-image / DALL-E / Sora)") == "OpenAI"
+        assert _vendor_of("DALL-E 3") == "OpenAI"
+
+    def test_google_variants(self):
+        assert _vendor_of("Google (Gemini / Imagen)") == "Google"
+        assert _vendor_of("Imagen 3") == "Google"
+
+    def test_other_vendors(self):
+        assert _vendor_of("Ideogram AI") == "Ideogram"
+        assert _vendor_of("Adobe Firefly") == "Adobe"
+        assert _vendor_of("Stability AI (Stable Image)") == "Stability AI"
+
+    def test_camera_label_is_not_an_ai_vendor(self):
+        # Camera platform labels must NOT normalize to an AI vendor, or a camera
+        # capture would be mistaken for AI-generation in clash detection.
+        assert _vendor_of("Leica (camera, C2PA capture)") is None
+
+    def test_unknown_is_none(self):
+        assert _vendor_of("a regular photo") is None
+        assert _vendor_of(None) is None
+
+
+class TestIntegrityClashesHelper:
+    def test_two_ai_vendors_clash(self):
+        clashes = _integrity_clashes({"c2pa": "OpenAI", "exif_generator": "Ideogram"}, None, camera_has_ai_marker=True)
+        assert len(clashes) == 1
+        assert "OpenAI" in clashes[0]
+        assert "Ideogram" in clashes[0]
+
+    def test_same_vendor_two_signals_no_clash(self):
+        # C2PA Google + SynthID-Google proxy is consistent, not a contradiction.
+        assert _integrity_clashes({"c2pa": "Google", "synthid": "Google"}, None, camera_has_ai_marker=True) == []
+
+    def test_single_vendor_no_clash(self):
+        assert _integrity_clashes({"c2pa": "OpenAI"}, None, camera_has_ai_marker=True) == []
+
+    def test_empty_no_clash(self):
+        assert _integrity_clashes({}, None, camera_has_ai_marker=False) == []
+
+    def test_camera_plus_ai_marker_clashes(self):
+        clashes = _integrity_clashes(
+            {"exif_generator": "Ideogram"},
+            "Google Pixel (camera, C2PA capture)",
+            camera_has_ai_marker=True,
+        )
+        assert any("Camera-capture" in c and "Pixel" in c for c in clashes)
+
+    def test_camera_without_ai_marker_no_clash(self):
+        # A clean camera capture (the normal case for our Pixel/Leica/Sony files)
+        # must NOT raise a clash.
+        assert _integrity_clashes({}, "Leica (camera, C2PA capture)", camera_has_ai_marker=False) == []
+
+
+class TestIntegrityClashEndToEnd:
+    def _c2pa_jpeg(self, tmp_path: Path, blob: bytes) -> Path:
+        path = tmp_path / "img.jpg"
+        path.write_bytes(b"\xff\xd8\xff\xe1jumbc2pa" + blob + b"\xff\xd9")
+        return path
+
+    def test_two_generator_stamps_clash(self, tmp_path: Path):
+        # An OpenAI C2PA manifest (AI source) on an image that ALSO carries a
+        # China TC260 AIGC label = two independent generator stamps naming
+        # different origins -> a laundering tell.
+        path = self._c2pa_jpeg(tmp_path, b"OpenAI ... trainedAlgorithmicMedia ... TC260:AIGC label")
+        r = identify(path, check_visible=False, check_invisible=False)
+        assert r.integrity_clashes
+        assert any("Conflicting AI-origin" in c for c in r.integrity_clashes)
+
+    def test_single_stamp_no_clash(self, tmp_path: Path):
+        path = self._c2pa_jpeg(tmp_path, b"OpenAI ... trainedAlgorithmicMedia")
+        r = identify(path, check_visible=False, check_invisible=False)
+        assert r.integrity_clashes == []
+
+    def test_clash_serializes_to_json(self, tmp_path: Path):
+        path = self._c2pa_jpeg(tmp_path, b"OpenAI ... trainedAlgorithmicMedia ... TC260:AIGC label")
+        r = identify(path, check_visible=False, check_invisible=False)
+        payload = json.loads(json.dumps(asdict(r), default=str))
+        assert payload["integrity_clashes"] == r.integrity_clashes
+
+
+@pytest.mark.skipif(not SAMPLES_DIR.exists(), reason="data/samples not present")
+@pytest.mark.parametrize("fixture", ["chatgpt-1.png", "firefly-1.png", "doubao-1.png", "grok-1.jpg", "mj-1.png"])
+class TestRealSamplesHaveNoClash:
+    """Every real single-origin fixture must report zero clashes (false-positive guard)."""
+
+    def test_no_false_positive_clash(self, fixture: str):
+        path = SAMPLES_DIR / fixture
+        if not path.exists():
+            pytest.skip(f"{fixture} not present")
+        r = identify(path, check_visible=False, check_invisible=False)
+        assert r.integrity_clashes == []

tests/test_image_io.py

@@ -0,0 +1,74 @@
+"""Tests for Unicode-safe cv2 image IO (issue #17)."""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import cv2
+import numpy as np
+
+from remove_ai_watermarks import image_io
+
+if TYPE_CHECKING:
+    from pathlib import Path
+
+# Non-ASCII filenames that break cv2.imread/imwrite on Windows (issue #17).
+_UNICODE_NAMES = [
+    "jimeng-2026-05-27-一面白色的墙.png",  # Chinese
+    "тест-изображение.png",  # Cyrillic
+    "café-señor.png",  # accented Latin
+]
+
+
+def _make_bgr() -> np.ndarray:
+    img = np.zeros((8, 8, 3), dtype=np.uint8)
+    img[2:6, 2:6] = (10, 120, 240)  # a BGR block so the round-trip is checkable
+    return img
+
+
+class TestUnicodeRoundTrip:
+    def test_write_then_read_preserves_pixels(self, tmp_path: Path) -> None:
+        for name in _UNICODE_NAMES:
+            path = tmp_path / name
+            src = _make_bgr()
+            assert image_io.imwrite(path, src) is True
+            assert path.exists()
+            out = image_io.imread(path)
+            assert out is not None
+            # PNG is lossless: pixels must match exactly.
+            assert np.array_equal(out, src)
+
+    def test_alpha_round_trip_with_unchanged_flag(self, tmp_path: Path) -> None:
+        path = tmp_path / "豆包-alpha.png"
+        bgra = np.zeros((8, 8, 4), dtype=np.uint8)
+        bgra[..., 3] = 128
+        assert image_io.imwrite(path, bgra) is True
+        out = image_io.imread(path, cv2.IMREAD_UNCHANGED)
+        assert out is not None
+        assert out.shape[2] == 4
+        assert np.array_equal(out, bgra)
+
+    def test_reads_file_written_by_raw_cv2(self, tmp_path: Path) -> None:
+        # An ASCII file written by plain cv2 must read back identically through
+        # the wrapper (decode path is byte-compatible with cv2.imread).
+        path = tmp_path / "ascii.png"
+        src = _make_bgr()
+        cv2.imwrite(str(path), src)
+        out = image_io.imread(path)
+        assert out is not None
+        assert np.array_equal(out, src)
+
+
+class TestFailureSemantics:
+    def test_missing_file_returns_none(self, tmp_path: Path) -> None:
+        assert image_io.imread(tmp_path / "does-not-exist-不存在.png") is None
+
+    def test_empty_file_returns_none(self, tmp_path: Path) -> None:
+        path = tmp_path / "empty.png"
+        path.write_bytes(b"")
+        assert image_io.imread(path) is None
+
+    def test_undecodable_file_returns_none(self, tmp_path: Path) -> None:
+        path = tmp_path / "garbage.png"
+        path.write_bytes(b"not an image")
+        assert image_io.imread(path) is None

tests/test_img2img_runner.py

@@ -14,7 +14,11 @@ from unittest.mock import Mock
 import pytest
 
 from remove_ai_watermarks.noai import img2img_runner
-from remove_ai_watermarks.noai.img2img_runner import run_img2img, run_img2img_with_mps_fallback
+from remove_ai_watermarks.noai.img2img_runner import (
+    run_differential_with_mps_fallback,
+    run_img2img,
+    run_img2img_with_mps_fallback,
+)
 
 _MPS_OOM = "MPS backend out of memory (MPS allocated: 17.21 GiB, max allowed: 20.13 GiB)"
 
@@ -106,6 +110,80 @@ class TestMpsFallback:
         reload_on_cpu.assert_not_called()
 
 
+class TestDifferentialMpsFallback:
+    """The protect-text (Differential Diffusion) path shares the MPS->CPU
+    fallback contract; mock ``run_differential`` so no torch/model is needed."""
+
+    def test_mps_error_reloads_on_cpu_and_retries(self, monkeypatch: pytest.MonkeyPatch):
+        sentinel = object()
+        inner = Mock(side_effect=[RuntimeError(_MPS_OOM), sentinel])
+        monkeypatch.setattr(img2img_runner, "run_differential", inner)
+        reload_on_cpu = Mock(return_value="cpu_pipe")
+
+        img, device = run_differential_with_mps_fallback(
+            load_pipeline=Mock(return_value="gpu_pipe"),
+            image=object(),
+            change_map=object(),
+            strength=0.05,
+            num_inference_steps=50,
+            guidance_scale=7.5,
+            generator="gen",
+            device="mps",
+            set_progress=lambda _m: None,
+            reload_on_cpu=reload_on_cpu,
+        )
+
+        assert (img, device) == (sentinel, "cpu")
+        reload_on_cpu.assert_called_once()
+        assert inner.call_count == 2
+        # Retry uses the reloaded CPU pipeline, device "cpu", and drops the MPS
+        # generator (generator=None) for deterministic CPU execution.
+        retry_args = inner.call_args_list[1].args
+        assert retry_args[0] == "cpu_pipe"
+        assert retry_args[6] is None  # generator
+        assert retry_args[7] == "cpu"  # device
+
+    def test_happy_path_returns_original_device_without_reload(self, monkeypatch: pytest.MonkeyPatch):
+        sentinel = object()
+        monkeypatch.setattr(img2img_runner, "run_differential", Mock(return_value=sentinel))
+        reload_on_cpu = Mock()
+
+        img, device = run_differential_with_mps_fallback(
+            load_pipeline=Mock(return_value="gpu_pipe"),
+            image=object(),
+            change_map=object(),
+            strength=0.05,
+            num_inference_steps=50,
+            guidance_scale=7.5,
+            generator="gen",
+            device="mps",
+            set_progress=lambda _m: None,
+            reload_on_cpu=reload_on_cpu,
+        )
+
+        assert (img, device) == (sentinel, "mps")
+        reload_on_cpu.assert_not_called()
+
+    def test_non_mps_runtime_error_propagates(self, monkeypatch: pytest.MonkeyPatch):
+        monkeypatch.setattr(img2img_runner, "run_differential", Mock(side_effect=RuntimeError("CUDA out of memory")))
+        reload_on_cpu = Mock()
+
+        with pytest.raises(RuntimeError, match="CUDA"):
+            run_differential_with_mps_fallback(
+                load_pipeline=Mock(return_value="gpu_pipe"),
+                image=object(),
+                change_map=object(),
+                strength=0.05,
+                num_inference_steps=50,
+                guidance_scale=7.5,
+                generator="gen",
+                device="mps",
+                set_progress=lambda _m: None,
+                reload_on_cpu=reload_on_cpu,
+            )
+        reload_on_cpu.assert_not_called()
+
+
 class TestRunImg2Img:
     def test_returns_first_image_from_pipeline_result(self):
         sentinel = object()

tests/test_metadata.py

@@ -631,6 +631,11 @@ _MP4_FTYP = b"\x00\x00\x00\x18ftypmp42\x00\x00\x00\x00mp42isom"
 _MP4_MDAT = b"\x00\x00\x00\x10mdat" + b"videodat"
 
 
+def _box(box_type: bytes, payload: bytes) -> bytes:
+    """Build a 32-bit-size ISOBMFF box: [size:4][type:4][payload]."""
+    return (8 + len(payload)).to_bytes(4, "big") + box_type + payload
+
+
 class TestVideoC2pa:
     """C2PA in MP4 (ISOBMFF) -- detect + strip, reusing the image box walker."""
 
@@ -654,6 +659,121 @@ class TestVideoC2pa:
         assert has_ai_metadata(out) is False
 
 
+class TestLateProvenanceBox:
+    """A C2PA / provenance box placed AFTER a large mdat (streaming / non-faststart
+    MP4) must still be detected -- the fixed first-MB scan would miss it."""
+
+    def _mp4_late_c2pa(self, tmp_path: Path, gap: int = 1_500_000) -> Path:
+        from remove_ai_watermarks.metadata import C2PA_UUID
+
+        big_mdat = _box(b"mdat", b"\x00" * gap)  # > 1 MB pushes the manifest past the scan window
+        manifest = C2PA_UUID + b"OpenAI jumbf c2pa ... trainedAlgorithmicMedia ..."
+        p = tmp_path / "stream.mp4"
+        p.write_bytes(_MP4_FTYP + big_mdat + _box(b"uuid", manifest))
+        return p
+
+    def test_scan_c2pa_region_finds_late_box(self, tmp_path: Path):
+        from remove_ai_watermarks.metadata import C2PA_UUID
+        from remove_ai_watermarks.noai.isobmff import scan_c2pa_region
+
+        region = scan_c2pa_region(self._mp4_late_c2pa(tmp_path))
+        assert C2PA_UUID in region
+        assert b"trainedAlgorithmicMedia" in region
+
+    def test_fixed_window_would_have_missed_it(self, tmp_path: Path):
+        # Documents the regression the box walk fixes: the manifest is beyond 1 MB.
+        from remove_ai_watermarks.metadata import C2PA_UUID
+
+        p = self._mp4_late_c2pa(tmp_path)
+        assert C2PA_UUID not in p.read_bytes()[: 1024 * 1024]
+
+    def test_scan_head_includes_late_box(self, tmp_path: Path):
+        from remove_ai_watermarks.metadata import C2PA_UUID, scan_head
+
+        assert C2PA_UUID in scan_head(self._mp4_late_c2pa(tmp_path))
+
+    def test_has_ai_metadata_detects_late_manifest(self, tmp_path: Path):
+        assert has_ai_metadata(self._mp4_late_c2pa(tmp_path)) is True
+
+    def test_scan_c2pa_region_non_isobmff_is_empty(self, tmp_path: Path):
+        from remove_ai_watermarks.noai.isobmff import scan_c2pa_region
+
+        p = tmp_path / "not.bin"
+        p.write_bytes(b"\x89PNG\r\n\x1a\n not an isobmff file")
+        assert scan_c2pa_region(p) == b""
+
+    def test_front_placed_manifest_still_detected(self, tmp_path: Path):
+        # Regression: a faststart MP4 (manifest before mdat) is unaffected.
+        from remove_ai_watermarks.metadata import C2PA_UUID
+
+        manifest = C2PA_UUID + b"OpenAI ... trainedAlgorithmicMedia ..."
+        p = tmp_path / "front.mp4"
+        p.write_bytes(_MP4_FTYP + _box(b"uuid", manifest) + _box(b"mdat", b"\x00" * 100))
+        assert has_ai_metadata(p) is True
+
+
+_AI_XMP = (
+    b'<?xpacket begin="\xef\xbb\xbf" id="W5M0MpCehiHzreSzNTczkc9d"?>'
+    b'<x:xmpmeta><TC260:AIGC>{"Label":"1"}</TC260:AIGC></x:xmpmeta>'
+    b'<?xpacket end="w"?>'
+)
+_PLAIN_XMP = (
+    b'<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?>'
+    b"<x:xmpmeta><dc:rights>(c) me</dc:rights></x:xmpmeta>"
+    b'<?xpacket end="w"?>'
+)
+
+
+class TestMetaBoxXmpBlanking:
+    """HEIF/AVIF store XMP as a meta-box ``mime`` item (bytes in mdat/idat), out of
+    reach of the top-level box stripper. An AI-label XMP packet there is blanked
+    in place (same length -> iloc offsets and image data stay intact)."""
+
+    def test_blanks_ai_packet_only(self):
+        from remove_ai_watermarks.noai.isobmff import blank_ai_xmp_packets
+
+        before, after = b"IMG_BEFORE" * 4, b"IMG_AFTER" * 4
+        data = before + _AI_XMP + after + _PLAIN_XMP
+        out, n = blank_ai_xmp_packets(data)
+        assert n == 1
+        assert len(out) == len(data)  # same length -> no offset shifts
+        assert b"TC260:AIGC" not in out  # AI label destroyed
+        assert before in out  # surrounding (image) bytes intact
+        assert after in out
+        assert b"dc:rights" in out  # plain XMP left alone
+
+    def test_no_packet_is_noop(self):
+        from remove_ai_watermarks.noai.isobmff import blank_ai_xmp_packets
+
+        data = b"just some mdat bytes, no xmp here"
+        assert blank_ai_xmp_packets(data) == (data, 0)
+
+    def test_plain_xmp_untouched(self):
+        from remove_ai_watermarks.noai.isobmff import blank_ai_xmp_packets
+
+        out, n = blank_ai_xmp_packets(_PLAIN_XMP)
+        assert n == 0
+        assert out == _PLAIN_XMP
+
+    def test_remove_ai_metadata_blanks_meta_box_xmp(self, tmp_path: Path):
+        # End-to-end: a HEIF with an AI XMP packet inside mdat is cleaned without
+        # touching the surrounding (coded image) bytes or the file length.
+        heic_ftyp = b"\x00\x00\x00\x18ftypheic\x00\x00\x00\x00heicmif1"
+        img = b"CODEDIMAGE" * 8
+        mdat = _box(b"mdat", img + _AI_XMP + img)
+        src = tmp_path / "ai.heic"
+        src.write_bytes(heic_ftyp + mdat)
+        assert has_ai_metadata(src) is True
+
+        out = tmp_path / "clean.heic"
+        remove_ai_metadata(src, out)
+        res = out.read_bytes()
+        assert len(res) == src.stat().st_size  # length preserved
+        assert b"TC260:AIGC" not in res
+        assert img in res  # coded image bytes intact
+        assert has_ai_metadata(out) is False
+
+
 class TestIsobmffMetadataRemoval:
     """Container-level AI-provenance stripping across ISOBMFF image/video/audio."""
 
@@ -718,9 +838,17 @@ class TestFfmpegMetadataStrip:
     def _wav_with_tag(self, path: Path, tag: str = "Suno AI") -> None:
         subprocess.run(  # noqa: S603
             [
-                shutil.which("ffmpeg"), "-y", "-loglevel", "error",
-                "-f", "lavfi", "-i", "sine=frequency=440:duration=0.1",
-                "-metadata", f"title={tag}", str(path),
+                shutil.which("ffmpeg"),
+                "-y",
+                "-loglevel",
+                "error",
+                "-f",
+                "lavfi",
+                "-i",
+                "sine=frequency=440:duration=0.1",
+                "-metadata",
+                f"title={tag}",
+                str(path),
             ],
             check=True,
         )

tests/test_text_protector.py

@@ -0,0 +1,59 @@
+"""Unit tests for the text-protection change-map helper (no model download).
+
+``build_change_map`` is the pure cv2/numpy part of ``text_protector``: it turns
+detected text polygons into a Differential-Diffusion change map. The polarity is
+load-bearing and was verified empirically (white = preserve, black = change), so
+a regression here would either freeze the whole image or fail to protect text.
+The PP-OCRv3 detector itself needs a model download and is not exercised here.
+"""
+
+from __future__ import annotations
+
+import numpy as np
+
+from remove_ai_watermarks.text_protector import build_change_map
+
+
+class TestBuildChangeMap:
+    def test_no_boxes_is_all_change(self):
+        m = build_change_map([], 32, 48)
+        assert m.shape == (32, 48)
+        assert m.dtype == np.float32
+        assert float(m.max()) == 0.0
+
+    def test_text_region_is_preserved_background_is_change(self):
+        # A 20x20 box centered in a 64x64 map, no feather for a crisp check.
+        box = np.array([[22, 22], [42, 22], [42, 42], [22, 42]])
+        m = build_change_map([box], 64, 64, preserve=0.9, feather=0)
+        # Inside the polygon: painted to preserve value.
+        assert m[32, 32] == np.float32(0.9)
+        # Far background: untouched -> full change (0.0).
+        assert m[2, 2] == 0.0
+        # Polarity: text preserved more than background.
+        assert m[32, 32] > m[2, 2]
+
+    def test_preserve_value_is_respected(self):
+        box = np.array([[10, 10], [30, 10], [30, 30], [10, 30]])
+        m = build_change_map([box], 40, 40, preserve=0.5, feather=0)
+        assert m[20, 20] == np.float32(0.5)
+
+    def test_feather_creates_soft_edge_gradient(self):
+        box = np.array([[20, 20], [44, 20], [44, 44], [20, 44]])
+        m = build_change_map([box], 64, 64, preserve=1.0, feather=15)
+        center = m[32, 32]
+        # An edge pixel just outside the polygon should be partially blended:
+        # strictly between full-change (0) and the preserved center.
+        edge = m[32, 47]
+        assert 0.0 < edge < center
+        assert center <= 1.0
+
+    def test_even_feather_does_not_crash(self):
+        box = np.array([[10, 10], [30, 10], [30, 30], [10, 30]])
+        m = build_change_map([box], 40, 40, feather=14)
+        assert m.shape == (40, 40)
+
+    def test_values_stay_in_unit_range(self):
+        box = np.array([[5, 5], [35, 5], [35, 35], [5, 35]])
+        m = build_change_map([box], 40, 40, preserve=1.0, feather=9)
+        assert float(m.min()) >= 0.0
+        assert float(m.max()) <= 1.0

uv.lock

@@ -1099,7 +1099,7 @@ wheels = [
 
 [[package]]
 name = "huggingface-hub"
-version = "1.16.4"
+version = "1.17.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "click" },
@@ -1113,9 +1113,9 @@ dependencies = [
     { name = "typer" },
     { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/ab/11/9b6e439cb2417c479c3da108b38363232a1554721de9f8ef4836cb07422b/huggingface_hub-1.16.4.tar.gz", hash = "sha256:023bacd155f837d3fa56379ac8e23dababe6d6d87b04f8dacc258a44a38abe01", size = 792585, upload-time = "2026-05-26T17:19:09.971Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/bd/65/9826515abb600b5722bcf53f8b4a2fb58340b1f8bfcaee19f83561c13a44/huggingface_hub-1.17.0.tar.gz", hash = "sha256:fad842b6763ef70ebc3919665b1b9273645203185400a7d6c5eddc2323cc3435", size = 797082, upload-time = "2026-05-28T15:12:13.347Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/da/86/e05d58ea272089151ba9f6fcc7b44a97aa2533d5a5bce46611220c23c6d6/huggingface_hub-1.16.4-py3-none-any.whl", hash = "sha256:994ec184c3330952d7b5f131ea0b1a6ba1047bd05461f5dec191f8fc1099fbd7", size = 668190, upload-time = "2026-05-26T17:19:08.228Z" },
+    { url = "https://files.pythonhosted.org/packages/02/28/d7cef5e477b855c25d415b8f57e5bc7347c7a90cad3acf1725d0c92ca294/huggingface_hub-1.17.0-py3-none-any.whl", hash = "sha256:3b8156d23118e87f6a587648bfbc04f04a12a757ccb4ed298b35c4ae638bf24c", size = 671546, upload-time = "2026-05-28T15:12:11.441Z" },
 ]
 
 [[package]]
@@ -2865,7 +2865,7 @@ wheels = [
 
 [[package]]
 name = "remove-ai-watermarks"
-version = "0.6.5"
+version = "0.6.10"
 source = { editable = "." }
 dependencies = [
     { name = "click" },
@@ -2892,6 +2892,7 @@ all = [
     { name = "pytest-cov" },
     { name = "ruff" },
     { name = "safetensors" },
+    { name = "tokenizers" },
     { name = "torch" },
     { name = "transformers" },
     { name = "trustmark" },
@@ -2913,6 +2914,7 @@ gpu = [
     { name = "controlnet-aux" },
     { name = "diffusers" },
     { name = "safetensors" },
+    { name = "tokenizers" },
     { name = "torch" },
     { name = "transformers" },
     { name = "ultralytics" },
@@ -2949,8 +2951,9 @@ requires-dist = [
     { name = "rich", specifier = ">=13.0.0" },
     { name = "ruff", marker = "extra == 'dev'", specifier = ">=0.4.0" },
     { name = "safetensors", marker = "extra == 'gpu'" },
+    { name = "tokenizers", marker = "extra == 'gpu'", specifier = ">=0.22,<0.23" },
     { name = "torch", marker = "extra == 'gpu'", specifier = ">=2.0.0" },
-    { name = "transformers", marker = "extra == 'gpu'", specifier = ">=4.35.0" },
+    { name = "transformers", marker = "extra == 'gpu'", specifier = ">=5,<6" },
     { name = "trustmark", marker = "extra == 'trustmark'", specifier = ">=0.8.0" },
     { name = "ultralytics", marker = "extra == 'gpu'", specifier = ">=8.0.0" },
 ]
@@ -3394,29 +3397,32 @@ wheels = [
 
 [[package]]
 name = "tokenizers"
-version = "0.23.0rc0"
+version = "0.22.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "huggingface-hub" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/0b/dc/2ba78324f6c82284f8d3d03bba16e5771d075aa4d5e9b4ecbd87af846af2/tokenizers-0.23.0rc0.tar.gz", hash = "sha256:685c6d269444451a2cf276d3f2bf655f3d7094be20c6553e413ede86b03c637b", size = 361629, upload-time = "2026-04-24T05:37:42.81Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/73/6f/f80cfef4a312e1fb34baf7d85c72d4411afde10978d4657f8cdd811d3ccc/tokenizers-0.22.2.tar.gz", hash = "sha256:473b83b915e547aa366d1eee11806deaf419e17be16310ac0a14077f1e28f917", size = 372115, upload-time = "2026-01-05T10:45:15.988Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/7f/b9/dda4065e0f4b62e0e5a625cbaeb928a611d847171e059066b3adfdb3866f/tokenizers-0.23.0rc0-cp310-abi3-macosx_10_12_x86_64.whl", hash = "sha256:bed69208ba6f74057e18e3c8ed73d62e681ff44f7be642ddeff747247c8a7a98", size = 3134709, upload-time = "2026-04-24T05:37:31.89Z" },
-    { url = "https://files.pythonhosted.org/packages/fa/16/54bd9f9e5c3641fe3d6d0e5b1cee37c58cb7520d22752c2065fc5a83caff/tokenizers-0.23.0rc0-cp310-abi3-macosx_11_0_arm64.whl", hash = "sha256:951be943c0657d8fd12e104731165a56d995c87533cd7f70a9444ddd7afa7708", size = 3043651, upload-time = "2026-04-24T05:37:30.305Z" },
-    { url = "https://files.pythonhosted.org/packages/86/11/54c1040ee93c8d74a364fbf4e17fd5d88e2eea940cbdba69d48d42a5a0c0/tokenizers-0.23.0rc0-cp310-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:704ffd50130f6c85aa76ad16c8218ff0f966b14c6e6cab7d0636e492e487ffa5", size = 3365683, upload-time = "2026-04-24T05:37:18.674Z" },
-    { url = "https://files.pythonhosted.org/packages/14/79/c8a7bdfee971346119349dab62f9918de512a7e5a8177555eaa50d854e1f/tokenizers-0.23.0rc0-cp310-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:bcd2a49117ad88999bc5d18d05addf67ec28e69f53e609ab07733c1f96404583", size = 3228688, upload-time = "2026-04-24T05:37:21.137Z" },
-    { url = "https://files.pythonhosted.org/packages/e1/32/a46ab1348d0b573dab69860eee601927b9934323e40f6f6018bb362a6013/tokenizers-0.23.0rc0-cp310-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:c52f927516521a3e1f6b6347f8bacedaf589eadd682e7ac87dac911d832c3a73", size = 3565137, upload-time = "2026-04-24T05:37:27.101Z" },
-    { url = "https://files.pythonhosted.org/packages/9c/f1/1a3b6a30388fe7d4b57b1ea7fcd6192341e479d65e50366ee0ba13d96d14/tokenizers-0.23.0rc0-cp310-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:1d6add82746146a6e052295ac429949c2d8e723244aa97ffe30cfee6cd788e98", size = 3826198, upload-time = "2026-04-24T05:37:22.783Z" },
-    { url = "https://files.pythonhosted.org/packages/a4/cb/161e52a424aa7ffb4097e8ce343d8dc2bdc42d590601032d4a9e6e5f7da5/tokenizers-0.23.0rc0-cp310-abi3-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:564115d3d6d2560b0a6b833d7dc39330d2328262557fbbd5bb0a14fb09b2b6cb", size = 3449011, upload-time = "2026-04-24T05:37:25.324Z" },
-    { url = "https://files.pythonhosted.org/packages/ff/31/0e4b77ca48b302a5db827584c9784f6cdbb35380c0dd1d7668712d477bb5/tokenizers-0.23.0rc0-cp310-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:82167864c62a3d83880ed23dea267aa5760e3fcf16fd73f94d413baf1968b211", size = 3337931, upload-time = "2026-04-24T05:37:28.723Z" },
-    { url = "https://files.pythonhosted.org/packages/50/e4/939249edee0073417b2c9447fd3b06e90c283ef6df72f3124427edae1f96/tokenizers-0.23.0rc0-cp310-abi3-manylinux_2_31_riscv64.whl", hash = "sha256:85f29751c4490bfaefe7e0d4b18ef28cd6d5f84c411e88ca896832eb4f18dd69", size = 3416560, upload-time = "2026-04-24T05:37:24.091Z" },
-    { url = "https://files.pythonhosted.org/packages/46/48/3a4bd2ba88af778e6fa6d03e271b2bc868f495745c8be91616781bf460d9/tokenizers-0.23.0rc0-cp310-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:f82b7578eaad0cbb72765d1fbaa7e7bc04c531337513a21f437b73e4617fcf46", size = 9810112, upload-time = "2026-04-24T05:37:33.679Z" },
-    { url = "https://files.pythonhosted.org/packages/45/8a/70c9919aefc7f514d6e98fb9be379b2850ca071a841d88900278781a07b0/tokenizers-0.23.0rc0-cp310-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:e61dff90a4ad8dc7e7e124d67756d63cf3ae57e32f04fb35bb408af91f47ea70", size = 9631038, upload-time = "2026-04-24T05:37:36.207Z" },
-    { url = "https://files.pythonhosted.org/packages/f9/f6/c15a5514f50bf953b70d3d2b7fd1829aa327ba8c9c519c54623510d6f459/tokenizers-0.23.0rc0-cp310-abi3-musllinux_1_2_i686.whl", hash = "sha256:5835b35d9a4815c8a4097d4dbac79c39b780684ea417fa4a93b9165e12ff1383", size = 9959195, upload-time = "2026-04-24T05:37:38.194Z" },
-    { url = "https://files.pythonhosted.org/packages/11/95/d1a6a0e6d6a9bc81b8124d83beb1fb1230310ee93938095f984a12fa336d/tokenizers-0.23.0rc0-cp310-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:33ed7df57a040ffb6f0244639619632a06f4c287ed1e77b5e70febb58f9e9a8b", size = 10106242, upload-time = "2026-04-24T05:37:40.745Z" },
-    { url = "https://files.pythonhosted.org/packages/78/c4/d9d587b9b32c9fca5ea901225d5c4c616802eb0082b17481d23808941641/tokenizers-0.23.0rc0-cp310-abi3-win32.whl", hash = "sha256:ab264a8ffdea05b5fd71a8bca6572762bde9b7aaadeba16dd25c7352a625fa71", size = 2523576, upload-time = "2026-04-24T05:37:47.173Z" },
-    { url = "https://files.pythonhosted.org/packages/d8/9b/34b36f6a47fec0a160887da23f173aa8a1729fa425ee67944c9be27f58de/tokenizers-0.23.0rc0-cp310-abi3-win_amd64.whl", hash = "sha256:27fe690eeb35a3a7e52f47d96c2ce8ffc6f939cc51a4591be86d2c86b9881267", size = 2788929, upload-time = "2026-04-24T05:37:45.81Z" },
-    { url = "https://files.pythonhosted.org/packages/35/ec/920d2b36ddddb5ce819a005d9650dc941935e534a27c48758c93388aaa5b/tokenizers-0.23.0rc0-cp310-abi3-win_arm64.whl", hash = "sha256:0b66c5eab2ddd26e59cfe6aa1945aa8b656ea0a9a715e24171c01b5ab1987630", size = 2655724, upload-time = "2026-04-24T05:37:44.108Z" },
+    { url = "https://files.pythonhosted.org/packages/92/97/5dbfabf04c7e348e655e907ed27913e03db0923abb5dfdd120d7b25630e1/tokenizers-0.22.2-cp39-abi3-macosx_10_12_x86_64.whl", hash = "sha256:544dd704ae7238755d790de45ba8da072e9af3eea688f698b137915ae959281c", size = 3100275, upload-time = "2026-01-05T10:41:02.158Z" },
+    { url = "https://files.pythonhosted.org/packages/2e/47/174dca0502ef88b28f1c9e06b73ce33500eedfac7a7692108aec220464e7/tokenizers-0.22.2-cp39-abi3-macosx_11_0_arm64.whl", hash = "sha256:1e418a55456beedca4621dbab65a318981467a2b188e982a23e117f115ce5001", size = 2981472, upload-time = "2026-01-05T10:41:00.276Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/84/7990e799f1309a8b87af6b948f31edaa12a3ed22d11b352eaf4f4b2e5753/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2249487018adec45d6e3554c71d46eb39fa8ea67156c640f7513eb26f318cec7", size = 3290736, upload-time = "2026-01-05T10:40:32.165Z" },
+    { url = "https://files.pythonhosted.org/packages/78/59/09d0d9ba94dcd5f4f1368d4858d24546b4bdc0231c2354aa31d6199f0399/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:25b85325d0815e86e0bac263506dd114578953b7b53d7de09a6485e4a160a7dd", size = 3168835, upload-time = "2026-01-05T10:40:38.847Z" },
+    { url = "https://files.pythonhosted.org/packages/47/50/b3ebb4243e7160bda8d34b731e54dd8ab8b133e50775872e7a434e524c28/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:bfb88f22a209ff7b40a576d5324bf8286b519d7358663db21d6246fb17eea2d5", size = 3521673, upload-time = "2026-01-05T10:40:56.614Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/fa/89f4cb9e08df770b57adb96f8cbb7e22695a4cb6c2bd5f0c4f0ebcf33b66/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:1c774b1276f71e1ef716e5486f21e76333464f47bece56bbd554485982a9e03e", size = 3724818, upload-time = "2026-01-05T10:40:44.507Z" },
+    { url = "https://files.pythonhosted.org/packages/64/04/ca2363f0bfbe3b3d36e95bf67e56a4c88c8e3362b658e616d1ac185d47f2/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:df6c4265b289083bf710dff49bc51ef252f9d5be33a45ee2bed151114a56207b", size = 3379195, upload-time = "2026-01-05T10:40:51.139Z" },
+    { url = "https://files.pythonhosted.org/packages/2e/76/932be4b50ef6ccedf9d3c6639b056a967a86258c6d9200643f01269211ca/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:369cc9fc8cc10cb24143873a0d95438bb8ee257bb80c71989e3ee290e8d72c67", size = 3274982, upload-time = "2026-01-05T10:40:58.331Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/28/5f9f5a4cc211b69e89420980e483831bcc29dade307955cc9dc858a40f01/tokenizers-0.22.2-cp39-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:29c30b83d8dcd061078b05ae0cb94d3c710555fbb44861139f9f83dcca3dc3e4", size = 9478245, upload-time = "2026-01-05T10:41:04.053Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/fb/66e2da4704d6aadebf8cb39f1d6d1957df667ab24cff2326b77cda0dcb85/tokenizers-0.22.2-cp39-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:37ae80a28c1d3265bb1f22464c856bd23c02a05bb211e56d0c5301a435be6c1a", size = 9560069, upload-time = "2026-01-05T10:45:10.673Z" },
+    { url = "https://files.pythonhosted.org/packages/16/04/fed398b05caa87ce9b1a1bb5166645e38196081b225059a6edaff6440fac/tokenizers-0.22.2-cp39-abi3-musllinux_1_2_i686.whl", hash = "sha256:791135ee325f2336f498590eb2f11dc5c295232f288e75c99a36c5dbce63088a", size = 9899263, upload-time = "2026-01-05T10:45:12.559Z" },
+    { url = "https://files.pythonhosted.org/packages/05/a1/d62dfe7376beaaf1394917e0f8e93ee5f67fea8fcf4107501db35996586b/tokenizers-0.22.2-cp39-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:38337540fbbddff8e999d59970f3c6f35a82de10053206a7562f1ea02d046fa5", size = 10033429, upload-time = "2026-01-05T10:45:14.333Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/18/a545c4ea42af3df6effd7d13d250ba77a0a86fb20393143bbb9a92e434d4/tokenizers-0.22.2-cp39-abi3-win32.whl", hash = "sha256:a6bf3f88c554a2b653af81f3204491c818ae2ac6fbc09e76ef4773351292bc92", size = 2502363, upload-time = "2026-01-05T10:45:20.593Z" },
+    { url = "https://files.pythonhosted.org/packages/65/71/0670843133a43d43070abeb1949abfdef12a86d490bea9cd9e18e37c5ff7/tokenizers-0.22.2-cp39-abi3-win_amd64.whl", hash = "sha256:c9ea31edff2968b44a88f97d784c2f16dc0729b8b143ed004699ebca91f05c48", size = 2747786, upload-time = "2026-01-05T10:45:18.411Z" },
+    { url = "https://files.pythonhosted.org/packages/72/f4/0de46cfa12cdcbcd464cc59fde36912af405696f687e53a091fb432f694c/tokenizers-0.22.2-cp39-abi3-win_arm64.whl", hash = "sha256:9ce725d22864a1e965217204946f830c37876eee3b2ba6fc6255e8e903d5fcbc", size = 2612133, upload-time = "2026-01-05T10:45:17.232Z" },
+    { url = "https://files.pythonhosted.org/packages/84/04/655b79dbcc9b3ac5f1479f18e931a344af67e5b7d3b251d2dcdcd7558592/tokenizers-0.22.2-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:753d47ebd4542742ef9261d9da92cd545b2cacbb48349a1225466745bb866ec4", size = 3282301, upload-time = "2026-01-05T10:40:34.858Z" },
+    { url = "https://files.pythonhosted.org/packages/46/cd/e4851401f3d8f6f45d8480262ab6a5c8cb9c4302a790a35aa14eeed6d2fd/tokenizers-0.22.2-pp310-pypy310_pp73-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:e10bf9113d209be7cd046d40fbabbaf3278ff6d18eb4da4c500443185dc1896c", size = 3161308, upload-time = "2026-01-05T10:40:40.737Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/6e/55553992a89982cd12d4a66dddb5e02126c58677ea3931efcbe601d419db/tokenizers-0.22.2-pp310-pypy310_pp73-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:64d94e84f6660764e64e7e0b22baa72f6cd942279fdbb21d46abd70d179f0195", size = 3718964, upload-time = "2026-01-05T10:40:46.56Z" },
+    { url = "https://files.pythonhosted.org/packages/59/8c/b1c87148aa15e099243ec9f0cf9d0e970cc2234c3257d558c25a2c5304e6/tokenizers-0.22.2-pp310-pypy310_pp73-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:f01a9c019878532f98927d2bacb79bbb404b43d3437455522a00a30718cdedb5", size = 3373542, upload-time = "2026-01-05T10:40:52.803Z" },
 ]
 
 [[package]]
@@ -3675,7 +3681,7 @@ wheels = [
 
 [[package]]
 name = "ultralytics"
-version = "8.4.55"
+version = "8.4.56"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "matplotlib", version = "3.10.9", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
@@ -3693,9 +3699,9 @@ dependencies = [
     { name = "torchvision" },
     { name = "ultralytics-thop" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/e9/0c/ec86cf04f520333fbf6fa4932a36e34c40588ddacb72091a8d352066ba25/ultralytics-8.4.55.tar.gz", hash = "sha256:27f0d5f9e818926ccf69da542cf53be0db896e659878bfc774fcd605f696a26a", size = 1065513, upload-time = "2026-05-26T17:26:05.265Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/38/73/440a7aa53aafaed46b106f1633385f390bebd300951cf15daf02a73b2c81/ultralytics-8.4.56.tar.gz", hash = "sha256:403f301cd4556b81fd13f5ca6dfcf2d3d6f31cf13cec0db79e655a086214318f", size = 1065497, upload-time = "2026-05-27T03:56:40.189Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/1c/59/914cd56ae9660fb1911f20560c505d7d672979912adceece1512506f149e/ultralytics-8.4.55-py3-none-any.whl", hash = "sha256:c2b2af4a531b8c7a93ef2ab66fe020bba5b30a64bd038f46f1fdaa3af8731fba", size = 1266451, upload-time = "2026-05-26T17:26:00.797Z" },
+    { url = "https://files.pythonhosted.org/packages/55/3f/7c86e841462a5f394c5f446943b5776b298aed418a6fd12386da70672660/ultralytics-8.4.56-py3-none-any.whl", hash = "sha256:84185b7c8ad91a8974bb4c5a42a14aeebb87a8c051cc64a5191758ed7d91fb95", size = 1266431, upload-time = "2026-05-27T03:56:35.698Z" },
 ]
 
 [[package]]