Add a Higgsfield DoP capture

Higgsfield's own DoP model ("Higgsfield Lite" in the web app, job model
image2video/lite) is not exposed by the MCP connector, so the capture was made
in the web app on a JPEG frame after a first job on a PNG failed and was
refunded. Like Genjutsu, the output carries no provenance: no C2PA, no TC260
label and no visible mark, only a Lavf re-mux.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

data/captures/2026-09/manifest.csv

@@ -133,3 +133,4 @@ af0265e41edec3413683f86441a1ffa9c64dbcb7117070d7ccbd2e91c20fd327,data/captures/2
 3f42e0877f67c4677f6f52be18146923fc5eeb471f9e0e27b9477d29dc1e6bbb,data/captures/2026-09/google-photos-web/forest-sky-vivid.jpg,new,google-photos-web,forest-sky-vivid.jpg,2026-09-26,,
 c83f903117cc030edf78164430ce59ffa897cb3a9efe464cf257e55644a955fd,data/captures/2026-09/google-photos-web/forest-crop-square.jpg,new,google-photos-web,forest-crop-square.jpg,2026-09-26,,
 e5612ea4bc06df48821d431d8d4758d3c73d1e74bd7b5b5dce921eaffdfbbe93,data/captures/2026-09/higgsfield/edit-kling-3-0-omni.mp4,new,higgsfield,edit-kling-3-0-omni.mp4,2026-09-26,True,Kuaishou Kling AI
+a76119db1b50f529e397875d38cf3938c5ae3b9e79d893d25567a7eee2340f41,data/captures/2026-09/higgsfield/dop-lite.mp4,new,higgsfield,dop-lite.mp4,2026-09-26,,

docs/watermarking-landscape.md

@@ -277,7 +277,7 @@ models used are paid-tier only.
 | Runway video models from five more vendors (2026-09-25): FLUX 3, Grok Imagine 1.5, Gemini Omni 1.1 Flash, HappyHorse 1.0, MiniMax H3 | None | Vendor provenance passed through: C2PA from Black Forest Labs, `xAI Grok Imagine` and Google (with its SynthID action), TC260 producers `001191330106MA2CFLDG4R10001` (Alibaba) and `MiniMax` | Vendor attributed on all five. The Omni clip was first reported as OpenAI Sora from a five-frame match on steam; Google's C2PA now vetoes it |
 | Higgsfield, 27 image models (Soul family, FLUX.2, GPT Image, Nano Banana, Seedream, Recraft, Grok Imagine, Kling O1, Wan 2.2, Z-Image) | None | `hf-job-id` PNG chunk on 23 of 26 PNGs, added after signing, so upstream C2PA (OpenAI, Black Forest Labs, BytePlus, Recraft, and fal.ai for Z-Image) fails `assertion.dataHash.mismatch`; Google and xAI manifests are stripped; Soul v1, Soul Cinema and Wan 2.2 carry nothing | Was reported as a Hugging Face job; now "Higgsfield (model not identified)" at medium confidence; the three bare files are unknown. Two files scored 0.53 as a Gemini sparkle on wood texture, now vetoed by their OpenAI or Kuaishou provenance (committed fixtures) |
 | Higgsfield video (Veo 3.1 and 3.1 Lite, Gemini Omni Flash and 1.1, Kling 3.0, 3.0 Turbo and 2.6, Grok Video and 1.5, Seedance 2.5, 2.0 and 1.5 Pro, MiniMax H3, H3 Max and Hailuo 2.3, Wan 2.5, 2.6, 2.7, 3.0 and 3.0 Prime, HappyHorse, FLUX.3 Video, Cinema Studio v2 and 3.0) | None | Google keeps C2PA with SynthID, xAI and Black Forest Labs keep C2PA, Seedance 2.x signs BytePlus C2PA; Kling, MiniMax and Alibaba write an MP4 `AIGC` TC260 label naming the producer as `kling`, `MiniMax` or Tongyi Yunqi's code (MiniMax with a `SecurityData` field); Higgsfield's own Cinema Studio v2 is Kling and 3.0 is Seedance by their provenance; Seedance 1.5 Pro has none | All attributed except Seedance 1.5 Pro (unknown). Veo 3.1 Lite was read as Kling, Cinema Studio v2 as a visible Kling mark and Hailuo 2.3 as a visible Hailuo mark, all on wood grain; each is now rejected (committed fixtures) |
-| Higgsfield own tools and video edits (Genjutsu motion control and object replacement, Marketing Studio; FLUX 3 Video Edit, Gemini Omni Flash 1.1 edit, Kling 3.0 Omni Edit, all on our own clips) | None | Genjutsu outputs carry no provenance; Marketing Studio signs BytePlus C2PA (Seedance); FLUX 3 edit records `compositeWithTrainedAlgorithmicMedia` with `opened` and `edited`; Gemini Omni edit keeps Google C2PA with SynthID; Kling Omni Edit writes a TC260 label with the bare producer `kling` and no C2PA (2026-09-26; it rejected a 360p source, and the Higgsfield connector passes the source as a reference image) | Genjutsu unknown; the others attributed. Genjutsu motion control was a visible Kling match on wood grain, now rejected by the Kling anchor requirement (committed fixture) |
+| Higgsfield own tools and video edits (Genjutsu motion control and object replacement, DoP (Higgsfield Lite), Marketing Studio; FLUX 3 Video Edit, Gemini Omni Flash 1.1 edit, Kling 3.0 Omni Edit, all on our own clips) | None | Genjutsu and DoP outputs carry no provenance (DoP: re-muxed by Lavf, no C2PA or TC260, 2026-09-26); Marketing Studio signs BytePlus C2PA (Seedance); FLUX 3 edit records `compositeWithTrainedAlgorithmicMedia` with `opened` and `edited`; Gemini Omni edit keeps Google C2PA with SynthID; Kling Omni Edit writes a TC260 label with the bare producer `kling` and no C2PA (2026-09-26; it rejected a 360p source, and the Higgsfield connector passes the source as a reference image) | Genjutsu unknown; the others attributed. Genjutsu motion control was a visible Kling match on wood grain, now rejected by the Kling anchor requirement (committed fixture) |
 | Dreamina, all 14 image models (Seedream 3.0 to 5.0 Pro, GPT Image 2 and 2.5, Nano Banana and Pro) and 8 video models (Seedance 1.0 to 2.5, MiniMax H3), paid plan | Boxed `AI` badge, top-left; not detected | ByteDance C2PA: active claim `ByteDance Media Transcode Service`, `Dreamina/7.5.0` in the ingredient; the OpenAI and Google manifests of hosted models are dropped | All read as ByteDance Dreamina, so the hosted model is not identifiable (committed fixtures) |
 
 Not captured, with the reason: Doubao (region block outside