Expand provenance metadata discovery

README.md

@@ -133,6 +133,8 @@ inspection includes the native TC260 `AIGC` tag in
 `moov.udta.meta.keys/ilst`, including a `moov` placed after the media payload,
 plus the QuickTime-form `meta` variants Doubao's iOS export writes (a bare
 `meta` box as a direct `moov` child, and a keyless `hdlr=mdir` metadata list).
+Keyed `workflow` and `prompt` entries in the same metadata-list structure are
+also detected and removed, including ComfyUI exports.
 MKV and WebM inspection reads the normative
 `Segment.Tags.Tag.SimpleTag` placement. AVI uses `LIST/INFO/AIGC`, while FLV
 uses `script.onMetaData.AIGC`. The non-ISOBMFF formats are remuxed with stream
@@ -350,9 +352,9 @@ Visible mark support includes:
 - one calibrated Samsung Galaxy AI label variant.
 
 Metadata and provenance inspection covers C2PA, EXIF, XMP, IPTC, common
-generator parameters, China TC260 AIGC labels, and several vendor specific
-signals. Optional decoders add support for open DWT-DCT watermarks and Adobe
-TrustMark.
+generator parameters in image and video containers, China TC260 AIGC labels,
+and several vendor specific signals. Optional decoders add support for open
+DWT-DCT watermarks and Adobe TrustMark.
 
 The exact support matrix, including important locale and detector limits, lives
 in [supported signals](docs/supported-signals.md). The
@@ -369,8 +371,9 @@ Visible removal follows three steps:
 
 Metadata removal uses format aware stripping. JPEG metadata removal preserves
 the encoded image scan instead of recompressing it. Native MP4/MOV TC260 values
-are blanked without changing box sizes or media offsets. Other supported
-containers use their corresponding metadata path.
+and keyed generation parameters are blanked without changing box sizes or
+media offsets. Other supported containers use their corresponding metadata
+path.
 
 Invisible removal is different. It regenerates the image through a diffusion
 pipeline to disrupt pixel and frequency domain watermarks. This changes the

docs/cli.md

@@ -316,10 +316,11 @@ delegates to the same verified metadata scanner and stripper as the generic
 streams are not transcoded. For MP4 and MOV, this includes the native TC260
 `AIGC` key and JSON value stored in `moov.udta.meta.keys/ilst`, plus the
 QuickTime-form `meta` variants Doubao's iOS export writes (a bare `meta` box
-as a direct `moov` child, and a keyless `hdlr=mdir` metadata list). The inspector
-seeks past a large `mdat` to find a tail `moov`. Removal stream-copies the
+as a direct `moov` child, and a keyless `hdlr=mdir` metadata list). Keyed
+`workflow` and `prompt` entries in the same metadata list are also inspected.
+The inspector seeks past a large `mdat` to find a tail `moov`. Removal stream-copies the
 container in bounded chunks, converts supported top-level provenance boxes to
-same-size `free` boxes, and blanks the TC260 key/value in place. Box sizes,
+same-size `free` boxes, and blanks the TC260 or generation key/value in place. Box sizes,
 media offsets, and encoded stream bytes do not move; the result is atomically
 published only after the complete copy succeeds.
 

docs/module-internals.md

@@ -222,7 +222,9 @@ alone is spaced out). The MP4/MOV/M4V/M4A removal path first validates the top-l
 box walk, then copies the source to a sibling temporary file in bounded chunks.
 Supported C2PA/JUMBF/AI-label boxes become same-size `free` boxes with blank
 payloads; TC260 removal changes the four-byte key to `free` and blanks only the
-validated JSON value with same-length spaces. This preserves every box size,
+validated JSON value with same-length spaces. Keyed generation metadata such
+as `workflow` and `prompt` uses the same structural walk; removal blanks the
+complete key and value in place. This preserves every box size,
 `stco`/`co64` offset, encoded stream byte, and source-sized memory bound.
 Publication is atomic, and a malformed top-level walk is copied unchanged. A
 generic `AIGC` key whose value has no TC260 field is ignored.
@@ -541,10 +543,18 @@ Key contracts:
   asserting that the pixels were generated; explicit `aigc_info` discriminator
   values and Dreamina `exportType=generation` do assert AI origin. Ordinary
   Aweme, retouch, and `lv` editor exports are preserved.
+- EXIF reads fall back to Pillow's decoded `getexif()` mapping when a container
+  exposes no ordinary `info["exif"]` blob. This covers ImageMagick PNG raw EXIF
+  profiles and keeps portable-record identification aligned with the file path.
+- Structured `generationParams.modelName` JSON is reduced to its recognized
+  model name before attribution, so prompts and the surrounding JSON never
+  become a platform label.
 - ISOBMFF containers use
   [`_internal/isobmff.py`](../src/remove_ai_watermarks/_internal/isobmff.py).
 - Native MP4/MOV TC260 `AIGC` entries are read from
   `moov.udta.meta.keys/ilst` and blanked without changing box sizes.
+- Native MP4/MOV `workflow` and `prompt` entries use that keyed metadata-list
+  reader and the same offset-preserving removal path.
 - Native MKV/WebM TC260 `AIGC` entries are read from
   `Segment.Tags.Tag.SimpleTag` and removed through the ffmpeg stream-copy path.
 - Native AVI and FLV TC260 entries are read from `LIST/INFO/AIGC` and

docs/python-api.md

@@ -568,7 +568,8 @@ field records that source audio is copied if present and not watermark-verified.
 MP4/MOV
 inspection recognizes the native TC260 `AIGC` entry in
 `moov.udta.meta.keys/ilst` and the QuickTime-form `meta` variants Doubao's iOS
-export writes; its removal preserves container size and encoded
+export writes. It also recognizes keyed `workflow` and `prompt` generation
+metadata in the same metadata-list structure; removal preserves container size and encoded
 stream bytes. MP4/MOV/M4V are copied in bounded chunks, so a large `mdat` is not
 loaded into memory; publication is atomic. MKV/WebM inspection recognizes the corresponding
 `Segment.Tags.Tag.SimpleTag` representation; its removal requires ffmpeg for a

docs/supported-signals.md

@@ -80,17 +80,20 @@ The learned backends download model files on first use.
 The inspection and stripping code handles signals in these groups:
 
 - C2PA Content Credentials and supported cloud manifest references;
-- EXIF and XMP generator fields;
+- EXIF and XMP generator fields, including structured
+  `generationParams.modelName` values from supported generators;
 - exact app-export provenance and AIGC disclosures from supported
   ByteDance-family products, with product-only provenance excluded from the
   generated-image verdict;
 - IPTC AI disclosure fields;
-- PNG text chunks and embedded generation parameters;
+- PNG text chunks and embedded generation parameters, including keyed
+  `workflow` and `prompt` tags in MP4/MOV metadata lists;
 - China TC260 AIGC labels in supported image placements and the normative
   MP4/MOV `moov.udta.meta.keys/ilst`, MKV/WebM
   `Segment.Tags.Tag.SimpleTag`, AVI `LIST/INFO/AIGC`, and FLV
   `script.onMetaData.AIGC` placements;
-- xAI and Grok EXIF signature fields;
+- xAI and Grok EXIF signature fields in ordinary EXIF and ImageMagick PNG raw
+  EXIF profiles;
 - Samsung AI editing markers;
 - Hugging Face job metadata;
 - open Stable Diffusion style DWT-DCT watermarks with the `detect` extra;

docs/watermarking-landscape.md

@@ -27,6 +27,9 @@ provider. For AI claims, exact product mappings in `claim_generator` therefore
 take precedence over issuer attribution. Supported mappings include Higgsfield
 AI, Recraft, Topaz Labs Image API, and TikTok Ad Creative Toolbox; an unknown
 claim generator still falls back to the certificate issuer.
+Exact claim-generator mappings also cover `Grok Imagine` as xAI Grok Imagine
+and `Suno` as Suno. Both can carry a valid AI source claim without a separately
+readable issuer, so issuer-only attribution leaves the platform blank.
 
 **OpenRouter raster metadata survey (2026-09-02):** a paid Recraft V3 API
 output carried a valid C2PA claim with `claim_generator` `recraft.ai`, a
@@ -37,11 +40,11 @@ container metadata. Their absence is evidence about those exact files and API
 routes, not a claim that every Krea or Riverflow export is metadata-free.
 
 **ByteDance Volcano Engine (Volcengine)** — the cloud behind Doubao / Jimeng — signs its AI image output with a cert from `certificate_center@volcengine.com` + `trainedAlgorithmicMedia` (issuer `b"volcengine"` → "ByteDance (Volcano Engine)", platform "ByteDance Volcano Engine"); note this is the C2PA-signed surface, distinct from the XMP/PNG TC260 `AIGC` label Doubao also uses. ByteDance's **international brand (BytePlus / Seedream / Seededit)** signs the same content as **"Byteplus Pte. Ltd."**. The bare `volcengine` needle missed it, so BytePlus output was mis-attributed to "Adobe Firefly" through an incidental "Adobe XMP" toolkit string. Issuer `b"Byteplus"` maps directly to "BytePlus (ByteDance)". ByteDance's consumer app **Dreamina** (the international Jimeng brand) signs as **"Bytedance Pte. Ltd."** with a `Dreamina/x.y` claim generator but, unlike the Volcano Engine surface, ships **no `trainedAlgorithmicMedia`**. Issuer `b"Dreamina"` maps to "ByteDance Dreamina" with **`asserts_ai=True`**. Registering the broader **issuer** `b"Bytedance Pte"` was deliberately avoided because that same entity also signs non-AI CapCut edits; keying on the `Dreamina` generator token is precise.
-- **EXIF/XMP/PNG-text generator tag (caught by `exif_generator`):** **Ideogram** writes EXIF `Make="Ideogram AI"` (collected 2026-05-24 — no C2PA, no SynthID, no imwatermark; the Make tag is the only signal). Additional verified generator stamps include **NovelAI** (`Software`, `Source`, and `Title` PNG text chunks), **Reve** (`Software` or XMP `CreatorTool` = `reve.com`), and **Aphrodite AI** (`Make` or `Software` = `Aphrodite AI`).
+- **EXIF/XMP/PNG-text generator tag (caught by `exif_generator`):** **Ideogram** writes EXIF `Make="Ideogram AI"` (collected 2026-05-24 — no C2PA, no SynthID, no imwatermark; the Make tag is the only signal). Additional verified generator stamps include **NovelAI** (`Software`, `Source`, and `Title` PNG text chunks), **Reve** (`Software` or XMP `CreatorTool` = `reve.com`), and **Aphrodite AI** (`Make` or `Software` = `Aphrodite AI`). A JSON `generationParams.modelName` value is parsed structurally and reduced to a recognized model name, rather than exposing its surrounding prompt JSON as a platform label.
 - **App-export provenance and AIGC JSON:** supported ByteDance-family exports can place a JSON object in EXIF `ImageDescription` or `UserComment`, independently of C2PA or TC260. Exact `product` values for Doubao, Xinghui, and Dreamina are removable product provenance, but do not alone prove that the pixels were generated. Dreamina additionally requires `exportType=generation` for that verdict. A nested Aweme `aigc_type=1` or private ByteDance `aigc_label_type=1` / `2` is an AIGC disclosure; `0` is inconclusive and can occur on a Dreamina generation export. Plain Aweme, retouch, and `lv` exports are preserved. The lower-case private field is deliberately not interpreted as the normative TC260 `Label`, whose values `1` / `2` / `3` mean generated / possibly generated / suspected generated under [GB 45438-2025](https://www.tc260.org.cn/upload/2025-03-15/1742009439794081593.pdf).
-- **xAI / Grok — its own EXIF signature scheme, NOT C2PA (DETECTED by `metadata.xai_signature`, built 2026-05-26).**
+- **xAI / Grok legacy JPEG EXIF signature scheme (DETECTED by `metadata.xai_signature`, built 2026-05-26).**
 
-Grok JPEG downloads (Aurora model) carry **no C2PA, no XMP, no SynthID, no IPTC** — only EXIF `Artist` = a UUID and EXIF `ImageDescription` = `Signature: <base64>` (a crypto signature, unverifiable locally without xAI's public key). This empirically kills the earlier unverified "xAI signs C2PA as xAI" lead — xAI is not even a C2PA member. `exif_generator` misses it (neither field holds an `AI_GENERATOR_TOKENS` token), so a dedicated detector `xai_signature(path)` matches the pair (`ImageDescription ~ ^Signature: [A-Za-z0-9+/=]{64,}` AND UUID `Artist`); wired into `has_ai_metadata`, `get_ai_metadata` (key `xai_signature`), and `identify` (signal `xai_signature`, platform "xAI (Grok / Aurora)").
+The sampled Grok JPEG downloads from the Aurora route carry **no C2PA, no XMP, no SynthID, no IPTC** — only EXIF `Artist` = a UUID and EXIF `ImageDescription` = `Signature: <base64>` (a crypto signature, unverifiable locally without xAI's public key). This is route-specific, not a product-wide no-C2PA claim: newer Grok Imagine output can carry a valid AI-source C2PA claim whose `claim_generator` is `Grok Imagine`. `exif_generator` misses the legacy pair (neither field holds an `AI_GENERATOR_TOKENS` token), so a dedicated detector `xai_signature(path)` matches it (`ImageDescription ~ ^Signature: [A-Za-z0-9+/=]{64,}` AND UUID `Artist`); wired into `has_ai_metadata`, `get_ai_metadata` (key `xai_signature`), and `identify` (signal `xai_signature`, platform "xAI (Grok / Aurora)").
 
 **Format confirmed stable across n=3 genuine generations:** exactly three EXIF tags (`Artist`, `ExifOffset`, `ImageDescription`), `Signature:` prefix constant, base64 payload 300-1004 chars. Two capture facts: (a) the `Artist` UUID **equals the public image id** in the asset URL (`https://imagine-public.x.ai/imagine-public/images/<uuid>.jpg`), so it is NOT a private per-user secret — only the `Signature` blob is; (b) the Grok web-UI image is a re-encoded **WebP with no signature** — the EXIF survives only in the *original* JPEG (download button or that public tokenless URL), which is why screenshots / re-encodes are metadata-stripped. A real fixture `data/fixtures/provenance/grok-1.jpg` plus **synthetic** JPEG fixtures (fake UUID + fake `Signature:` blob) cover the detector; never add a real Grok image carrying private content (the repo is public).
 
@@ -49,7 +52,10 @@ Grok JPEG downloads (Aurora model) carry **no C2PA, no XMP, no SynthID, no IPTC*
 JPEG EXIF, which deletes the xAI Signature and UUID Artist pair plus supported
 AI generator values while retaining unrelated camera and editor EXIF. The
 shared `xai_signature_pair` helper is the single source of truth for the
-pair. On the ISOBMFF path, `blank_ai_exif_tokens` provides the corresponding
+pair. Converted PNGs can retain the same pair in an ImageMagick raw EXIF
+profile; Pillow exposes that form through `getexif()` even when `info["exif"]`
+is absent, so detection, removal, and portable records share that fallback. On
+the ISOBMFF path, `blank_ai_exif_tokens` provides the corresponding
 in-place scrub for supported EXIF values, TC260 AIGC blocks, and the xAI pair.
 - **China TC260 AIGC label (caught by `AIGC_MARKERS` / `metadata.aigc_label`, surfaced by `identify` as the `aigc` signal):** China-served generators embed an XMP `<TC260:AIGC>{"Label":"1","ContentProducer":...}` block — China's mandatory AI-content labeling (TC260 namespace `tc260.org.cn/ns/AIGC`). The label says only "this is AI", but its `ContentProducer` names the signing entity — `001` + `1` + an 18-char Unified Social Credit Code + a 5-digit product suffix, normalized by `metadata.uscc_of`, or for a few generators a bare product name. `KnownMark.tc260_producer_codes` maps the codes settled per vendor by `scripts/vendor_cohort_harvest.py` to registry mark keys, so an AIGC image relaxes only the detector of the product it actually carries; an unmapped or absent producer leaves every product strict. The same mapping supplies the manufacturer for provenance-conflict checks. A code identifies a legal entity, not necessarily one brand, so a hosting or aggregating platform that signs for several apps is a recall bet rather than a proof.
 
@@ -88,7 +94,10 @@ payloads. Removal remuxes either container through ffmpeg with stream copy.
 
 Native TC260 MP4/MOV tags do not live in those top-level provenance boxes.
 `tc260_aigc_payloads` separately seeks through `moov.udta.meta.keys/ilst`, so
-the normative tag is also found when a large `mdat` precedes `moov`.
+the normative tag is also found when a large `mdat` precedes `moov`. The parser
+canonicalizes known fields across PascalCase and lowerCamelCase serializations.
+The same keyed metadata-list walk exposes generation `workflow` and `prompt`
+entries, including ComfyUI exports, and blanks them in place during removal.
 
 **Meta-box XMP and EXIF removal are handled in place:** an AI-label XMP packet
 stored as a meta-box `mime` item is blanked by

scripts/corpus_gap_scan.py

@@ -29,6 +29,7 @@ from __future__ import annotations
 
 import csv
 import logging
+import re
 from collections import Counter
 from datetime import date
 from importlib.metadata import version
@@ -40,7 +41,14 @@ from _plain_console import Console, Table
 
 from remove_ai_watermarks.identify import _metadata_region as identify_metadata_region
 from remove_ai_watermarks.identify import identify
-from remove_ai_watermarks.metadata import scan_head
+from remove_ai_watermarks.metadata import (
+    IPTC_AI_FIELD_MARKERS,
+    IPTC_AI_MARKERS,
+    aigc_label_from_metadata,
+    c2pa_marker_in,
+    samsung_genai_in,
+    scan_head,
+)
 
 log = logging.getLogger(__name__)
 console = Console()
@@ -51,22 +59,10 @@ console = Console()
 # below is the primary guard, this list is the second. Group: C2PA/JUMBF infra,
 # AI source-type / labeling schemes, and distinctive generator name strings.
 MARKERS: tuple[bytes, ...] = (
-    # C2PA / JUMBF infrastructure and AI source-type / labeling schemes.
-    b"c2pa",
-    b"jumbf",
-    b"contentauth",
+    # AI source-type / labeling schemes that are meaningful by presence.
     b"trainedAlgorithmicMedia",
-    b"digitalSourceType",
-    b'"AIGC"',
-    b"<TC260:AIGC>",
-    b"TC260:AIGC",
-    b"tc260.org.cn",
     b"AISystemUsed",
-    b"SynthID",
     b"hf-job-id",
-    b"genAIType",
-    b"PhotoEditor_Re_Edit",
-    b"Signature:",
     # Distinctive multi-word generator strings only. Bare single words (Luma,
     # Gemini, Sora, ...) are omitted: they collide with unrelated metadata prose
     # (e.g. "Luma" in Lightroom's EnhanceDenoiseLumaAmount), defeating precision.
@@ -80,12 +76,11 @@ MARKERS: tuple[bytes, ...] = (
     b"Adobe Firefly",
     b"Black Forest",
     b"volcengine",
-    b"Doubao",
-    b"\xe8\xb1\x86\xe5\x8c\x85",
     b"Nano Banana",
     b"Stability AI",
-    b"Samsung Galaxy",
 )
+_XAI_SIGNATURE_RE = re.compile(rb"Signature:\s*[A-Za-z0-9+/=]{64,}")
+_UUID_RE = re.compile(rb"[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}", re.IGNORECASE)
 
 REPORT_FIELDS: tuple[str, ...] = (
     "path",
@@ -122,9 +117,20 @@ def _row(rep, *, path: str, suffix: str, lib_version: str) -> dict[str, str]:  #
 
 
 def _marker_hits(region: bytes) -> list[str]:
-    """Return marker labels found case-insensitively in one metadata region."""
+    """Return high-precision marker labels found in one metadata region."""
     folded = region.lower()
-    return sorted({marker.decode("latin-1", "replace") for marker in MARKERS if marker.lower() in folded})
+    hits = {marker.decode("latin-1", "replace") for marker in MARKERS if marker.lower() in folded}
+    if c2pa_marker_in(region):
+        hits.add("C2PA/JUMBF")
+    if aigc_label_from_metadata(region) is not None:
+        hits.add("TC260 AIGC")
+    if any(marker in region for marker in IPTC_AI_MARKERS + IPTC_AI_FIELD_MARKERS):
+        hits.add("IPTC AI disclosure")
+    if samsung_genai_in(region) is not None:
+        hits.add("Samsung genAIType")
+    if _XAI_SIGNATURE_RE.search(region) and _UUID_RE.search(region):
+        hits.add("xAI signature pair")
+    return sorted(hits)
 
 
 def _marker_hits_for_path(path: Path) -> list[str]:

src/remove_ai_watermarks/_internal/constants.py

@@ -119,6 +119,8 @@ C2PA_IDENTITY_AI_ORGS = frozenset(vendor.org for vendor in C2PA_AI_VENDORS if ve
 # Keep this attribution beside the issuer registry so every C2PA consumer has one
 # canonical source rather than maintaining a derived product map in identify.py.
 C2PA_CLAIM_GENERATOR_PLATFORMS: tuple[tuple[str, str], ...] = (
+    ("grok imagine", "xAI Grok Imagine"),
+    ("suno", "Suno"),
     ("adobe_firefly", "Adobe Firefly"),
     ("firefly", "Adobe Firefly"),
     ("dreamina", "ByteDance Dreamina"),
@@ -211,6 +213,7 @@ AI_GENERATOR_TOKENS = frozenset(
         "invokeai",
         "imagen",
         "gpt-image",
+        "nano banana",
         "nightcafe",
         "ideogram",
         "leonardo",

src/remove_ai_watermarks/_internal/isobmff.py

@@ -43,6 +43,7 @@ from remove_ai_watermarks.metadata import (
     C2PA_UUID,
     IPTC_AI_FIELD_MARKERS,
     IPTC_AI_MARKERS,
+    LOCAL_GENERATION_METADATA_KEYS,
     MAX_TC260_VALUE_BYTES,
     parse_tc260_aigc_json,
 )
@@ -152,12 +153,12 @@ def iter_file_boxes(
         pos = box_end
 
 
-def _tc260_key_indices(
+def _metadata_key_indices(
     stream: BinaryIO,
     payload_off: int,
     box_end: int,
-) -> dict[int, tuple[int, int]]:
-    """Map every exact ``AIGC`` key index to its byte span."""
+) -> dict[int, tuple[bytes, tuple[int, int]]]:
+    """Map bounded metadata key indices to their names and byte spans."""
     if payload_off + 8 > box_end:
         return {}
     stream.seek(payload_off)
@@ -166,7 +167,7 @@ def _tc260_key_indices(
         return {}
     entry_count = struct.unpack(">I", prefix[4:8])[0]
     pos = payload_off + 8
-    found: dict[int, tuple[int, int]] = {}
+    found: dict[int, tuple[bytes, tuple[int, int]]] = {}
     for index in range(1, entry_count + 1):
         if pos + 8 > box_end:
             return {}
@@ -179,10 +180,12 @@ def _tc260_key_indices(
         if entry_size < 8 or entry_end > box_end:
             return {}
         name_start = pos + 8
-        if entry_end - name_start == 4:
+        name_size = entry_end - name_start
+        if 0 < name_size <= 128:
             stream.seek(name_start)
-            if stream.read(4) == b"AIGC":
-                found[index] = (name_start, entry_end)
+            name = stream.read(name_size)
+            if len(name) == name_size:
+                found[index] = (name, (name_start, entry_end))
         pos = entry_end
     return found
 
@@ -248,12 +251,23 @@ def _tc260_aigc_regions(
     iOS export writes, where the JSON sits in a bare ``ilst`` data item with no
     key name to blank.
     """
-    regions: list[tuple[tuple[int, int] | None, int, int, bytes]] = []
+    return [
+        (key_span, value_start, value_end, value)
+        for key_name, key_span, value_start, value_end, value in _metadata_items(stream, file_size)
+        if (key_name is None or key_name == b"AIGC") and parse_tc260_aigc_json(value) is not None
+    ]
+
+
+def _metadata_items(
+    stream: BinaryIO,
+    file_size: int,
+) -> Iterator[tuple[bytes | None, tuple[int, int] | None, int, int, bytes]]:
+    """Yield bounded keyed or keyless metadata-list values from a media file."""
     for _moov_start, moov_end, moov_type, moov_payload in iter_file_boxes(stream, 0, file_size):
         if moov_type != b"moov":
             continue
         for meta_payload, meta_end in _iter_tc260_meta_boxes(stream, moov_payload, moov_end):
-            keys: dict[int, tuple[int, int]] = {}
+            keys: dict[int, tuple[bytes, tuple[int, int]]] = {}
             ilst_boxes: list[tuple[int, int]] = []
             keyed = False
             for _child_start, child_end, child_type, child_payload in _meta_child_boxes(
@@ -263,7 +277,7 @@ def _tc260_aigc_regions(
             ):
                 if child_type == b"keys":
                     keyed = True
-                    keys.update(_tc260_key_indices(stream, child_payload, child_end))
+                    keys.update(_metadata_key_indices(stream, child_payload, child_end))
                 elif child_type == b"ilst":
                     ilst_boxes.append((child_payload, child_end))
             if not ilst_boxes:
@@ -275,8 +289,8 @@ def _tc260_aigc_regions(
                     ilst_end,
                 ):
                     index = int.from_bytes(item_type, "big")
-                    key_span = keys.get(index)
-                    if keyed and key_span is None:
+                    key = keys.get(index)
+                    if keyed and key is None:
                         # A keyed (ISO) meta box maps items through ``keys``;
                         # an unmapped index is not an AIGC entry, and reading
                         # its value would pull arbitrary metadata (e.g. cover
@@ -284,6 +298,7 @@ def _tc260_aigc_regions(
                         # keyless QuickTime list falls through to content
                         # validation below.
                         continue
+                    key_name, key_span = key if key is not None else (None, None)
                     for _data_start, data_end, data_type, data_payload in iter_file_boxes(
                         stream,
                         item_payload,
@@ -295,9 +310,8 @@ def _tc260_aigc_regions(
                             continue
                         stream.seek(value_start)
                         value = stream.read(value_size)
-                        if len(value) == value_size and parse_tc260_aigc_json(value) is not None:
-                            regions.append((key_span, value_start, data_end, value))
-    return regions
+                        if len(value) == value_size:
+                            yield key_name, key_span, value_start, data_end, value
 
 
 def tc260_aigc_payloads(path: str | Path, *, strict: bool = False) -> tuple[bytes, ...]:
@@ -315,6 +329,42 @@ def tc260_aigc_payloads(path: str | Path, *, strict: bool = False) -> tuple[byte
         return ()
 
 
+def _ai_metadata_regions(
+    stream: BinaryIO,
+    file_size: int,
+) -> list[tuple[tuple[int, int], int, int, str, bytes]]:
+    """Locate keyed generation metadata without reading media payloads."""
+    regions: list[tuple[tuple[int, int], int, int, str, bytes]] = []
+    for name_bytes, key_span, value_start, value_end, value in _metadata_items(stream, file_size):
+        if name_bytes is None or key_span is None:
+            continue
+        try:
+            name = name_bytes.decode("utf-8")
+        except UnicodeDecodeError:
+            continue
+        if name.lower() in LOCAL_GENERATION_METADATA_KEYS:
+            regions.append((key_span, value_start, value_end, name, value))
+    return regions
+
+
+def ai_metadata_tags(path: str | Path, *, strict: bool = False) -> dict[str, str]:
+    """Read keyed AI-generation metadata from an ISOBMFF container."""
+    try:
+        with open(path, "rb") as stream:
+            if not is_isobmff(stream.read(8)):
+                return {}
+            stream.seek(0, 2)
+            file_size = stream.tell()
+            return {
+                name: value.decode("utf-8", "replace")
+                for _key_span, _value_start, _value_end, name, value in _ai_metadata_regions(stream, file_size)
+            }
+    except OSError:
+        if strict:
+            raise
+        return {}
+
+
 def blank_tc260_aigc_tags(data: bytes) -> tuple[bytes, int]:
     """Blank native TC260 values in place while preserving every box offset.
 
@@ -576,7 +626,9 @@ def strip_isobmff_media_file(
             max_scan=max_box_scan,
         )
         tc260_regions = _tc260_aigc_regions(stream, file_size) if targets is not None else []
+        ai_metadata_regions = _ai_metadata_regions(stream, file_size) if targets is not None else []
         tc260_key_spans = {region[0] for region in tc260_regions if region[0] is not None}
+        ai_metadata_key_spans = {region[0] for region in ai_metadata_regions}
 
     with atomic_video_output(output_path) as temporary_path:
         with source_path.open("rb") as source_stream, temporary_path.open("r+b") as temporary:
@@ -591,6 +643,10 @@ def strip_isobmff_media_file(
                         temporary.seek(key_span[0])
                         temporary.write(b"free")
                     _overwrite_range(temporary, value_start, value_end, byte=b" ")
+                for key_span, value_start, value_end, _name, _value in ai_metadata_regions:
+                    temporary.seek(key_span[0])
+                    temporary.write(b"free" + b" " * (key_span[1] - key_span[0] - 4))
+                    _overwrite_range(temporary, value_start, value_end, byte=b" ")
             temporary.flush()
             os.fsync(temporary.fileno())
         shutil.copymode(source_path, temporary_path)
@@ -601,7 +657,7 @@ def strip_isobmff_media_file(
             source_path,
         )
         return 0, 0
-    return len(targets), len(tc260_key_spans)
+    return len(targets), len(tc260_key_spans | ai_metadata_key_spans)
 
 
 def strip_c2pa_boxes(data: bytes) -> tuple[bytes, int]:

src/remove_ai_watermarks/identify.py

@@ -51,6 +51,7 @@ from remove_ai_watermarks.metadata import (
     AIGC_MARKERS,
     IPTC_AI_FIELD_MARKERS,
     IPTC_AI_MARKERS,
+    LOCAL_GENERATION_METADATA_KEYS,
     aigc_label,
     aigc_label_from_metadata,
     c2pa_cloud_manifest_in,
@@ -106,9 +107,7 @@ _ISSUER_PLATFORM: tuple[tuple[str, str], ...] = tuple(
 
 # PNG-text / EXIF keys that indicate a local diffusion pipeline (vs. a hosted
 # platform's C2PA). Subset of AI_METADATA_KEYS; excludes the C2PA/Software keys.
-_LOCAL_GEN_KEYS = frozenset(
-    AI_METADATA_KEYS & {"parameters", "prompt", "negative_prompt", "workflow", "comfyui", "invokeai_metadata", "dream"}
-)
+_LOCAL_GEN_KEYS = LOCAL_GENERATION_METADATA_KEYS
 
 _STRIP_CAVEAT = (
     "Absence of metadata is not proof the image is clean: C2PA, EXIF, and PNG "

src/remove_ai_watermarks/metadata.py

@@ -59,6 +59,9 @@ AI_METADATA_KEYS: frozenset[str] = frozenset(
         "Software",
     ]
 )
+LOCAL_GENERATION_METADATA_KEYS: frozenset[str] = frozenset(
+    AI_METADATA_KEYS & {"parameters", "prompt", "negative_prompt", "workflow", "comfyui", "invokeai_metadata", "dream"}
+)
 
 AI_KEYWORDS: tuple[str, ...] = (
     "stable_diffusion",
@@ -169,6 +172,7 @@ TC260_AIGC_FIELDS: frozenset[str] = frozenset(
         "ServiceUser",
     }
 )
+_TC260_CANONICAL_FIELDS: dict[str, str] = {field.casefold(): field for field in TC260_AIGC_FIELDS}
 MAX_TC260_VALUE_BYTES = 1024 * 1024
 
 
@@ -197,7 +201,10 @@ def parse_tc260_aigc_json(value: bytes) -> dict[str, str] | None:
         return None
     if not isinstance(parsed, dict):
         return None
-    fields = {str(key): str(item) for key, item in cast("dict[object, object]", parsed).items()}
+    fields = {
+        _TC260_CANONICAL_FIELDS.get(str(key).casefold(), str(key)): str(item)
+        for key, item in cast("dict[object, object]", parsed).items()
+    }
     return fields if TC260_AIGC_FIELDS & fields.keys() else None
 
 
@@ -877,11 +884,47 @@ def generator_from_metadata(candidates: Iterable[str], scan: bytes = b"") -> str
     for value in itertools.chain(candidates, creator_tools):
         if app_generator := app_generator_from_metadata(value):
             return app_generator
+        if structured_generator := _structured_generator_from_metadata(value):
+            return structured_generator
         if any(token in value.lower() for token in AI_GENERATOR_TOKENS):
             return value.strip()
     return None
 
 
+def _structured_generator_from_metadata(value: str) -> str | None:
+    """Extract a bounded generator name without returning its surrounding JSON."""
+    from remove_ai_watermarks._internal.constants import AI_GENERATOR_TOKENS
+
+    if len(value) > MAX_TC260_VALUE_BYTES:
+        return None
+    try:
+        parsed = json.loads(value)
+    except ValueError:
+        return None
+    if not isinstance(parsed, dict):
+        return None
+    generation_params = cast("dict[object, object]", parsed).get("generationParams")
+    if not isinstance(generation_params, dict):
+        return None
+    model_name = cast("dict[object, object]", generation_params).get("modelName")
+    if not isinstance(model_name, str) or not model_name.strip() or len(model_name) > 200:
+        return None
+    normalized = model_name.strip()
+    return normalized if any(token in normalized.lower() for token in AI_GENERATOR_TOKENS) else None
+
+
+def exif_bytes_from_image(image: Any) -> bytes | None:
+    """Return EXIF bytes from either a normal EXIF field or a decoded raw profile."""
+    value = image.info.get("exif")
+    if isinstance(value, bytes) and value:
+        return value
+    with contextlib.suppress(Exception):
+        exif = image.getexif()
+        if exif:
+            return exif.tobytes()
+    return None
+
+
 _APP_PROVENANCE_PRODUCTS: dict[str, str] = {
     "doubao": "ByteDance Doubao",
     "xinghui": "ByteDance Xinghui",
@@ -894,12 +937,17 @@ _APP_AIGC_LABEL_RE = re.compile(r'"aigc_label_type"\s*:\s*[12](?=\s*[,}])', re.I
 _APP_AIGC_TYPE_RE = re.compile(r'"aigc_type"\s*:\s*1(?=\s*[,}])', re.IGNORECASE)
 
 
-def _normalized_app_metadata(value: str | bytes) -> str:
-    text = value.decode("latin-1", "ignore") if isinstance(value, bytes) else value
+def _normalized_app_metadata(value: object) -> str:
+    if isinstance(value, bytes):
+        text = value.decode("latin-1", "ignore")
+    elif isinstance(value, str):
+        text = value
+    else:
+        return ""
     return text.replace('\\"', '"')
 
 
-def _app_metadata_evidence(value: str | bytes) -> tuple[str | None, str | None]:
+def _app_metadata_evidence(value: object) -> tuple[str | None, str | None]:
     """Return removable product provenance and stronger AI-origin evidence."""
     normalized = _normalized_app_metadata(value)
     products = tuple(match.group(1).lower() for match in _APP_PRODUCT_RE.finditer(normalized))
@@ -962,7 +1010,7 @@ def exif_generator(image_path: Path) -> str | None:
 
         with Image.open(image_path) as img:
             info = img.info
-            exif_bytes = info.get("exif")
+            exif_bytes = exif_bytes_from_image(img)
             # PNG tEXt/iTXt chunks land in img.info too (same idiom as the other
             # PNG-text readers in this module); NovelAI stamps Software/Source/Title.
             for key in ("Software", "Source", "Title", "Description"):
@@ -1034,7 +1082,7 @@ def _xai_signature_impl(image_path: Path) -> bool:
         from PIL import Image
 
         with Image.open(image_path) as img:
-            exif_bytes = img.info.get("exif")
+            exif_bytes = exif_bytes_from_image(img)
         if not exif_bytes:
             return False
         tags = piexif.load(exif_bytes).get("0th", {})
@@ -1063,8 +1111,7 @@ def _is_aigc_exif_value(raw: object) -> bool:
         return False
     if b"AIGC" not in raw:
         return False
-    text = bytes(raw).decode("latin-1", "ignore")
-    return any(field in text for field in TC260_AIGC_FIELDS)
+    return aigc_label_from_metadata(bytes(raw)) is not None
 
 
 def _ai_exif_targets(loaded: dict[str, Any]) -> list[tuple[str, int, bytes, str]]:
@@ -1100,15 +1147,18 @@ def _ai_exif_targets(loaded: dict[str, Any]) -> list[tuple[str, int, bytes, str]
     if xai_signature_pair(_exif_text(ifd0, piexif.ImageIFD.ImageDescription), _exif_text(ifd0, piexif.ImageIFD.Artist)):
         add("0th", ifd0, piexif.ImageIFD.ImageDescription, "ImageDescription")
         add("0th", ifd0, piexif.ImageIFD.Artist, "Artist")
-    # (b) known AI generator token in a 0th text tag.
-    for tag, name in (
-        (piexif.ImageIFD.Software, "Software"),
-        (piexif.ImageIFD.Make, "Make"),
-        (piexif.ImageIFD.Artist, "Artist"),
-        (piexif.ImageIFD.ImageDescription, "ImageDescription"),
+    # (b) known AI generator token or structured generator name in a text tag.
+    for ifd_key, ifd, tag, name in (
+        ("0th", ifd0, piexif.ImageIFD.Software, "Software"),
+        ("0th", ifd0, piexif.ImageIFD.Make, "Make"),
+        ("0th", ifd0, piexif.ImageIFD.Artist, "Artist"),
+        ("0th", ifd0, piexif.ImageIFD.ImageDescription, "ImageDescription"),
+        ("Exif", ifde, piexif.ExifIFD.UserComment, "UserComment"),
     ):
-        if any(token in _exif_text(ifd0, tag).lower() for token in AI_GENERATOR_TOKENS):
-            add("0th", ifd0, tag, name)
+        value = ifd.get(tag)
+        text = value.decode("latin-1", "replace") if isinstance(value, bytes) else ""
+        if any(token in text.lower() for token in AI_GENERATOR_TOKENS) or _structured_generator_from_metadata(text):
+            add(ifd_key, ifd, tag, name)
     # (c) TC260 AIGC block in ImageDescription (0th) or UserComment (Exif sub-IFD).
     if _is_aigc_exif_value(ifd0.get(piexif.ImageIFD.ImageDescription)):
         add("0th", ifd0, piexif.ImageIFD.ImageDescription, "ImageDescription")
@@ -1171,6 +1221,11 @@ def get_ai_metadata(image_path: Path) -> dict[str, str]:
     except Exception as exc:
         logger.debug("PIL could not open %s for AI-metadata scan: %s", image_path, exc)
 
+    from remove_ai_watermarks._internal.isobmff import ai_metadata_tags
+
+    for key, value in ai_metadata_tags(image_path).items():
+        result.setdefault(key, value[:200] + ("…" if len(value) > 200 else ""))
+
     # C2PA manifest fields from the single canonical parser (_internal/c2pa.py).
     c2pa = extract_c2pa_info(image_path)
     for key in (
@@ -1632,11 +1687,11 @@ def remove_ai_metadata(
     with open(source_path, "rb") as f:
         head = f.read(26)
     if source_path.suffix.lower() in _STREAMING_ISOBMFF_EXTS and is_isobmff(head):
-        stripped, tc260_blanked = strip_isobmff_media_file(source_path, output_path)
+        stripped, native_ai_blanked = strip_isobmff_media_file(source_path, output_path)
         logger.info(
-            "Stream-blanked %d AI-provenance box(es) and %d native TC260 tag(s) → %s",
+            "Stream-blanked %d AI-provenance box(es) and %d native AI metadata tag(s) → %s",
             stripped,
-            tc260_blanked,
+            native_ai_blanked,
             output_path,
         )
         return output_path
@@ -1714,8 +1769,9 @@ def remove_ai_metadata(
         return output_path
 
     # Read image and filter metadata
-    with Image.open(source_path) as img:
-        img = img.copy()
+    with Image.open(source_path) as source_image:
+        source_exif_bytes = exif_bytes_from_image(source_image)
+        img = source_image.copy()
         # Pick the save format. Honor the caller's output extension (so a deliberate
         # source.png -> output.jpg conversion still works) UNLESS the SOURCE is misnamed
         # -- a lossless PNG/WebP whose extension lies (served as .jpg). There the output
@@ -1762,6 +1818,10 @@ def remove_ai_metadata(
             if _keep_standard_text_metadata(key, value, keep_standard):
                 kept_meta[key] = str(value) if not isinstance(value, str) else value
 
+        if source_exif_bytes:
+            with contextlib.suppress(Exception):
+                exif_data = piexif.load(source_exif_bytes)
+
         # Apply cleaned metadata
         if save_kwargs["format"] == "PNG" and kept_meta:
             pnginfo = PngInfo()

src/remove_ai_watermarks/metadata_record.py

@@ -50,6 +50,7 @@ from remove_ai_watermarks._internal.schema import require_schema_version
 from remove_ai_watermarks.metadata import (
     QUICK_SCAN_BYTES,
     SAMSUNG_EDITOR_MARKER,
+    exif_bytes_from_image,
     exif_text,
     read_file_tail,
 )
@@ -267,7 +268,10 @@ def _decoder_info(image_path: Path) -> dict[str, Any]:
         with Image.open(image_path) as img:
             # PIL types this mapping with a non-string key union (a DPI tuple key
             # exists), so the keys are normalized here rather than assumed.
-            return {str(key): value for key, value in img.info.items()}
+            info = {str(key): value for key, value in img.info.items()}
+            if "exif" not in info and (exif_bytes := exif_bytes_from_image(img)):
+                info["exif"] = exif_bytes
+            return info
     except (UnidentifiedImageError, ImportError) as exc:  # unsupported optional decoder
         logger.debug("PIL info unavailable for %s: %s", image_path, exc)
         return {}
@@ -392,6 +396,10 @@ def collect_metadata_record(
                 issues.append({"stage": "trailer", "code": "collection-failed"})
             try:
                 info = _decoder_info(image_path)
+                if container == "isobmff":
+                    from remove_ai_watermarks._internal.isobmff import ai_metadata_tags
+
+                    info.update(ai_metadata_tags(image_path, strict=True))
                 exif = _exif_pairs(info)
             except Exception as exc:
                 logger.debug("metadata decoder collection failed for %s: %s", image_path, exc)

tests/test_identify.py

@@ -464,6 +464,8 @@ class TestIdentifyNonPng:
     @pytest.mark.parametrize(
         ("claim_generator", "platform"),
         [
+            ("Grok Imagine", "xAI Grok Imagine"),
+            ("Suno", "Suno"),
             ("Higgsfield AI", "Higgsfield AI"),
             ("recraft.ai", "Recraft"),
             ("Topaz Labs Image API", "Topaz Labs"),

tests/test_metadata.py

@@ -945,6 +945,30 @@ class TestExifGenerator:
     def test_clean_image_is_none(self, tmp_clean_png: Path):
         assert exif_generator(tmp_clean_png) is None
 
+    def test_nested_generation_model_in_webp_user_comment(self, tmp_path: Path):
+        import json
+
+        payload = json.dumps(
+            {
+                "generationParams": {
+                    "modelName": "Nano Banana synthetic fixture",
+                    "prompt": "private prompt must not become the platform label",
+                }
+            }
+        )
+        exif = piexif.dump({"Exif": {piexif.ExifIFD.UserComment: payload.encode("utf-8")}})
+        source = tmp_path / "generated.webp"
+        Image.new("RGB", (32, 32)).save(source, "WEBP", lossless=True, exif=exif)
+
+        assert exif_generator(source) == "Nano Banana synthetic fixture"
+        assert (
+            identify(source, check_visible=False).platform == "Nano Banana synthetic fixture (embedded generator tag)"
+        )
+
+        output = tmp_path / "clean.webp"
+        remove_ai_metadata(source, output)
+        assert exif_generator(output) is None
+
 
 _FAKE_SIG = "A" * 120  # 64+ base64 chars; real Grok payloads are 300-1004
 _FAKE_UUID = "12345678-1234-1234-1234-123456789abc"
@@ -970,12 +994,48 @@ def _grok_jpeg(tmp_path: Path, *, signature: str = _FAKE_SIG, artist: str = _FAK
     return path
 
 
+def _grok_imagemagick_png(tmp_path: Path) -> Path:
+    """Write the ImageMagick raw-profile EXIF form seen on converted PNGs."""
+    exif = piexif.dump(
+        {
+            "0th": {
+                piexif.ImageIFD.ImageDescription: f"Signature: {_FAKE_SIG}".encode("latin1"),
+                piexif.ImageIFD.Artist: _FAKE_UUID.encode("latin1"),
+                piexif.ImageIFD.Make: b"Canon",
+            },
+            "Exif": {},
+            "GPS": {},
+            "1st": {},
+        }
+    )
+    profile = (
+        "\nexif\n"
+        + f"{len(exif):8d}\n"
+        + "\n".join(exif.hex()[start : start + 72] for start in range(0, len(exif) * 2, 72))
+    )
+    info = PngInfo()
+    info.add_text("Raw profile type exif", profile)
+    path = tmp_path / "grok-imagemagick.png"
+    Image.new("RGB", (64, 64), (70, 80, 90)).save(path, pnginfo=info)
+    return path
+
+
 class TestXaiSignature:
     """xAI / Grok's EXIF Signature + UUID-Artist provenance scheme."""
 
     def test_signature_plus_uuid_detected(self, tmp_path: Path):
         assert xai_signature(_grok_jpeg(tmp_path)) is True
 
+    def test_imagemagick_png_raw_exif_profile_detected(self, tmp_path: Path):
+        path = _grok_imagemagick_png(tmp_path)
+
+        with Image.open(path) as image:
+            assert "exif" not in image.info
+            assert image.getexif()
+
+        assert xai_signature(path) is True
+        assert "xai_signature" in get_ai_metadata(path)
+
     def test_real_grok_sample_detected(self):
         # Real committed Grok download (data/fixtures/provenance/grok-1.jpg); the EXIF
         # Signature + UUID-Artist pair is the only AI signal it carries.
@@ -1169,6 +1229,21 @@ class TestRemoveAiExif:
         assert xai_signature(out) is False
         assert has_ai_metadata(out) is False
 
+    def test_grok_raw_profile_stripped_on_png_output(self, tmp_path: Path):
+        src = _grok_imagemagick_png(tmp_path)
+        out = tmp_path / "clean.png"
+
+        remove_ai_metadata(src, out)
+
+        assert xai_signature(out) is False
+        with Image.open(out) as image:
+            assert image.getexif().get(piexif.ImageIFD.Make) == "Canon"
+
+    def test_non_text_exif_value_does_not_break_app_metadata_scrub(self):
+        from remove_ai_watermarks.metadata import _app_metadata_evidence
+
+        assert _app_metadata_evidence((1, 2)) == (None, None)
+
     def test_generator_make_token_stripped(self, tmp_path: Path):
         # Ideogram's EXIF Make="Ideogram AI" must be scrubbed on removal.
         exif = piexif.dump({"0th": {piexif.ImageIFD.Make: b"Ideogram AI"}, "Exif": {}, "GPS": {}, "1st": {}})
@@ -1391,6 +1466,37 @@ class TestAIGCLabel:
         assert info["Label"] == "1"
         assert info["ContentProducer"] == "001191440300708461136T1308L"
 
+    def test_parses_lower_camel_case_tc260_fields(self, tmp_path: Path):
+        import json
+
+        from remove_ai_watermarks.metadata import aigc_label
+
+        p = tmp_path / "aigc-lower-camel.jpg"
+        payload = json.dumps(
+            {
+                "data": {
+                    "AIGC": {
+                        "label": "1",
+                        "contentProducer": "synthetic-producer",
+                        "produceId": "synthetic-id",
+                    }
+                }
+            }
+        )
+        exif = {"Exif": {piexif.ExifIFD.UserComment: payload.encode("ascii")}}
+        Image.new("RGB", (32, 32)).save(p, exif=piexif.dump(exif))
+
+        info = aigc_label(p)
+
+        assert info == {
+            "Label": "1",
+            "ContentProducer": "synthetic-producer",
+            "ProduceID": "synthetic-id",
+        }
+        out = tmp_path / "clean-lower-camel.jpg"
+        remove_ai_metadata(p, out)
+        assert aigc_label(out) is None
+
     def test_has_ai_metadata_detects_raw_json_exif_form(self, tmp_path: Path):
         assert has_ai_metadata(self._aigc_exif_jpeg(tmp_path))
 

tests/test_metadata_record.py

@@ -15,6 +15,7 @@ from pathlib import Path
 import numpy as np
 import pytest
 from PIL import Image
+from PIL.PngImagePlugin import PngInfo
 
 from remove_ai_watermarks.identify import (
     PROVENANCE_REPORT_SCHEMA_VERSION,
@@ -157,6 +158,54 @@ class TestPlacementsTheRecordCouldDrop:
         assert "xai_signature" in {s.name for s in identify(path, check_visible=False, check_invisible=False).signals}
         _assert_same_verdict(path)
 
+    def test_an_imagemagick_raw_exif_profile_survives(self, tmp_path: Path):
+        """Pillow decodes this PNG EXIF profile through getexif, not img.info."""
+        import piexif
+
+        exif = piexif.dump(
+            {
+                "0th": {
+                    piexif.ImageIFD.ImageDescription: b"Signature: " + b"A" * 80,
+                    piexif.ImageIFD.Artist: b"3f2504e0-4f89-11d3-9a0c-0305e82c3301",
+                }
+            }
+        )
+        profile = (
+            "\nexif\n"
+            + f"{len(exif):8d}\n"
+            + "\n".join(exif.hex()[start : start + 72] for start in range(0, len(exif) * 2, 72))
+        )
+        info = PngInfo()
+        info.add_text("Raw profile type exif", profile)
+        path = tmp_path / "grok-converted.png"
+        Image.fromarray(np.zeros((64, 64, 3), dtype=np.uint8)).save(path, "PNG", pnginfo=info)
+
+        assert "xai_signature" in {s.name for s in identify(path, check_visible=False, check_invisible=False).signals}
+        _assert_same_verdict(path)
+
+    def test_isobmff_generation_tags_survive(self, tmp_path: Path):
+        def box(kind: bytes, payload: bytes) -> bytes:
+            return (8 + len(payload)).to_bytes(4, "big") + kind + payload
+
+        def key(name: bytes) -> bytes:
+            return box(b"mdta", name)
+
+        def value(index: int, payload: bytes) -> bytes:
+            return box(index.to_bytes(4, "big"), box(b"data", b"\x00" * 8 + payload))
+
+        keys = box(b"keys", b"\x00" * 4 + (2).to_bytes(4, "big") + key(b"workflow") + key(b"title"))
+        ilst = box(b"ilst", value(1, b'{"app":"ComfyUI"}') + value(2, b"standard title"))
+        meta = box(b"meta", b"\x00" * 4 + keys + ilst)
+        path = tmp_path / "workflow.mp4"
+        path.write_bytes(
+            b"\x00\x00\x00\x18ftypmp42\x00\x00\x00\x00mp42isom"
+            + box(b"mdat", b"synthetic-video")
+            + box(b"moov", box(b"udta", meta))
+        )
+
+        assert "gen_params" in {s.name for s in identify(path, check_visible=False, check_invisible=False).signals}
+        _assert_same_verdict(path)
+
 
 class TestRecordShape:
     def test_the_record_survives_json(self, tmp_path: Path):

tests/test_signal_discovery_scripts.py

@@ -25,7 +25,16 @@ def _report(**overrides):
 
 
 def test_gap_markers_are_case_insensitive():
-    assert "c2pa" in gap._marker_hits(b"UPPERCASE C2PA MANIFEST")
+    assert "ComfyUI" in gap._marker_hits(b"COMFYUI WORKFLOW")
+
+
+def test_gap_markers_require_structural_provenance_evidence():
+    noise = b"c2pa digitalSourceType Samsung Galaxy Doubao PhotoEditor_Re_Edit Signature: " + b"A" * 80
+    assert gap._marker_hits(noise) == []
+
+    tc260 = b'{"AIGC":{"label":"1","contentProducer":"synthetic"}}'
+    assert "TC260 AIGC" in gap._marker_hits(tc260)
+    assert "Samsung genAIType" in gap._marker_hits(b'PhotoEditor_Re_Edit_Data{"genAIType":1}')
 
 
 def test_gap_candidates_cover_more_than_blind_unknowns():

tests/test_video.py

@@ -65,6 +65,30 @@ def _video_with_tc260(path: Path, *, media_payload: bytes = _VIDEO_PAYLOAD) -> P
     return path
 
 
+def _video_with_comfyui_metadata(path: Path, *, media_payload: bytes = _VIDEO_PAYLOAD) -> Path:
+    workflow = b'{"nodes":[{"type":"KSampler"}],"app":"ComfyUI"}'
+    prompt = b'{"1":{"class_type":"KSampler"}}'
+    keys = _box(
+        b"keys",
+        b"\x00\x00\x00\x00"
+        + (4).to_bytes(4, "big")
+        + _metadata_key(b"workflow")
+        + _metadata_key(b"prompt")
+        + _metadata_key(b"title")
+        + _metadata_key(b"software"),
+    )
+    ilst = _box(
+        b"ilst",
+        _metadata_value(1, workflow)
+        + _metadata_value(2, prompt)
+        + _metadata_value(3, b"standard title")
+        + _metadata_value(4, b"ordinary editor"),
+    )
+    meta = _box(b"meta", b"\x00\x00\x00\x00" + keys + ilst)
+    path.write_bytes(_MP4_FTYP + _box(b"mdat", media_payload) + _box(b"moov", _box(b"udta", meta)))
+    return path
+
+
 def _hdlr(handler: bytes) -> bytes:
     # version/flags + pre_defined + handler_type + reserved[3]; the TC260 walker
     # never parses it, but a real-shaped hdlr keeps the fixture honest.
@@ -564,6 +588,24 @@ class TestVideoDependencies:
 
 
 class TestVideoMetadataApi:
+    def test_inspects_and_removes_comfyui_mp4_metadata(self, tmp_path: Path):
+        from remove_ai_watermarks.video import inspect_video_metadata, remove_video_metadata
+
+        source = _video_with_comfyui_metadata(tmp_path / "source.mp4")
+        output = tmp_path / "clean.mp4"
+
+        report = inspect_video_metadata(source)
+        assert set(report.markers) >= {"workflow", "prompt"}
+
+        result = remove_video_metadata(source, output)
+        cleaned = output.read_bytes()
+        assert result.remaining == {}
+        assert len(cleaned) == source.stat().st_size
+        assert _VIDEO_PAYLOAD in cleaned
+        assert b"standard title" in cleaned
+        assert b"ordinary editor" in cleaned
+        assert b"ComfyUI" not in cleaned
+
     def test_top_level_api_is_lazy_exported(self):
         import remove_ai_watermarks as raiw