Skip to content

Add twelve com.originalpictures soft-binding algorithms: five watermarks, seven fingerprints (entries 68-79) - #104

Merged
domguinard merged 14 commits into
mainfrom
add-originalpictures-soft-bindings
Nov 16, 2026
Merged

domguinard merged 14 commits into
mainfrom
add-originalpictures-soft-bindings

Conversation

@drmahdikazempour

Copy link
Copy Markdown
Contributor

Moves #98 into a branch of this repository, as requested in that review (the fork could not be merged from). The entries are the ones the group approved in #98, including the suggestions applied there. Merged with current main; since main now ends at 67, the twelve entries are renumbered 68-79. No other change. Schema validation (jsonschema, as in this repo's workflow) passes on the merged list; identifiers and alg values are unique.


Adds twelve entries for Original Pictures Technologies, Inc., which runs these algorithms in its Content Credentials signing service: five watermarks and seven fingerprints. I work at Original Pictures and maintain these deployments. (This replaces #97, which was opened from a personal fork by mistake; the file is byte-identical.)

Watermarks

Proposed id alg Media What it is
68 com.originalpictures.pixelseal.1 image Meta FAIR PixelSeal (videoseal 870ca7fb). Our 61-bit identifier, repeated 4 times with a CRC-12 in 256 bits.
69 com.originalpictures.pixelseal-video.1 video PixelSeal in video mode. Our 51-bit identifier, repeated 5 times with parity in 256 bits.
70 com.originalpictures.audioseal.1 audio Meta FAIR AudioSeal 0.2.0 streaming. 16-bit identifier, our detection rule.
71 com.originalpictures.videoseal.1 video Meta FAIR VideoSeal 1.0. 51-bit identifier, repeat-5 with parity.
72 com.originalpictures.videoseal-image.1 image VideoSeal 1.0 in image mode. 61-bit identifier, repeat-4 with CRC-12.

Fingerprints (type: "fingerprint")

Proposed id alg Media What it is
73 com.originalpictures.dinov2.1 image Meta FAIR DINOv2 ViT-S/14: 384-dimension L2-normalised CLS descriptor of the whole image at 224x224; cosine ≥ 0.957.
74 com.originalpictures.pdq.1 image Meta PDQ (ThreatExchange baefb4ed): 256-bit hash, queried through all eight dihedral variants; Hamming ≤ 34.
75 com.originalpictures.dinov2-lsh256.1 image DINOv2 ViT-S/14 descriptor projected by a per-account keyed 384x256 Gaussian matrix to 256 sign bits; Hamming ≤ 19.
76 com.originalpictures.chromaprint.1 audio Chromaprint 1.6.1 (AcoustID), raw sub-fingerprints of the first 120 s; aligned bit score ≥ 0.75.
77 com.originalpictures.audfprint.1 audio audfprint by Dan Ellis (dpwe/audfprint cb03ba99): 11,025 Hz, density 20, 20-bit landmark hashes; ≥ 20 hashes on one offset.
78 com.originalpictures.tmk-pdqf.1 video Meta TMK+PDQF (ThreatExchange baefb4ed), default configuration (15 fps, 64x64); level-1 and level-2 ≥ 0.70.
79 com.originalpictures.vpdq.1 video Meta vPDQ (ThreatExchange baefb4ed): PDQ of one frame per second; ≥ 40% of quality-50 frames within Hamming 31.

Each fingerprint entry is a specific deployment of a published algorithm: the parameters, calibration and threshold above are ours, so a value is only comparable with values computed the same way. Each description and page says which parameters those are. We do not file our ISCC codes (already io.iscc.v0, 3) or a SHA-256 entry (com.joinmonolith.sha256, 48, already covers a generic content hash).

The identifiers are the next free numbers after the current main (max 67). Please renumber if another entry lands first.

Why not reuse 30, 31, 32, 50 or 54. Those entries use the same Meta FAIR watermark models, but each one is another company's implementation. Each has its own payload layout and resolves its identifiers through its own service. A decoder for 31 cannot read our PixelSeal payload, and our records are not in AIWatermark's resolver. Using their identifiers would point a validator at the wrong format and the wrong resolver. Each watermark description names the entries it differs from.

No softBindingResolutionApis. We have no public Soft Binding Resolution API yet, and the field is optional. Matching is done by our own verification, within the account that signed the file. We will open an amendment if we publish one.

Each informationalUrl is a page for that one algorithm, listed at https://originalpictures.com/soft-bindings/, with its parameters, thresholds and the value form a soft-binding assertion carries. Contact for all twelve: hello@originalpictures.com. The schema check (jsonschema validate, the same check as this repo's workflow) passes on the merged list.

drmahdikazempour and others added 14 commits November 9, 2026 12:45
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
Co-authored-by: Dominique Guinard <dom@guinard.org>
main now ends at 67 (com.art-koffee.phash.1), which the previous numbering (67-78) collided with. Only identifiers change; the entries are otherwise as reviewed in #98, including the suggestions applied there.
@drmahdikazempour

Copy link
Copy Markdown
Contributor Author

@domguinard as requested in #98, the change is now in a branch of this repository (add-originalpictures-soft-bindings) instead of our fork, so you can merge it directly.

  • Same twelve entries the group approved on Nov 11, including your id suggestions, which are carried over as commits.
  • Conflict resolved: merged with current main. Your suggestions allocated 67-78, but main has since assigned 67 to com.art-koffee.phash.1 (Add com.art-koffee.phash.1 (ArtKoffee perceptual image fingerprint) #99), so ours are now 68-79, in the same order. Only the identifiers changed.
  • The schema validation passes on the merged list, and identifiers and alg values are unique. GitHub reports it mergeable.

#98 is closed in favour of this one, and you've also been invited as a collaborator on our fork. Could you approve and merge when you have a moment? Thanks a lot.

@domguinard domguinard left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approving this as it has already been reviewed by the WM TF under: #98

@domguinard
domguinard merged commit 0c3eea7 into main Nov 16, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants