[DamaizLabs]

This PR adds an entry for the watermarks of Lanternist (https://github.com/DamaizLabs/lanternist), an open-source, MIT-licensed engine that generates faceless videos locally. I am its maintainer.

Lanternist falls under Article 50(2) of the EU AI Act, and follows Measure 3.4 (c) of the Code of Practice on marking and labelling of AI-generated content: every output carries a publicly readable signpost that tells which detector to use. Following the signpost mode described in adobe/trustmark/c2pa/README.md, a TrustMark Q watermark (BCH_SUPER) carries the identifier of this entry, next to the Video Seal and AudioSeal watermarks it points to.

A transparent note on the selection rules: Lanternist's watermarks carry a fixed, public message that identifies content generated by Lanternist. They do not carry a per-asset identifier, and Lanternist runs no manifest repository, because everything runs on the user's PC. The C2PA manifest declares them with c2pa.watermarked.unbound. If the list should only hold algorithms used for manifest recovery, I would be grateful for guidance on the right place to register a signpost target.

The informational URL documents the exact messages, thresholds and how to read each watermark with its authors' open-source decoders. The identifier 59 is requested because 55-57 are claimed by open PRs (#85, #87, #88); I am happy to take any number you assign. The updated list validates against softbinding-algorithm-list.schema.json (Draft 2020-12, with format checks).

🤖 Generated with Claude Code

https://claude.ai/code/session_01UNvaiTsw6zLckiMXyrTD7i

[jcollomosse]

This requires some discussion at the next TF meeting - the PR states that watermark is not a soft binding technology i.e. does not include a unique identifier... so should it belong on this list? I see the motivation in listing is in part to obtain a unique identifier code for the purposes of signposting, but signposting is an optional part of the soft binding lookup process, so it seems this obtaining such an identifier is not consistent with the purpose of the list. @domguinard

[domguinard]

Indeed let's discuss this today @jcollomosse

[domguinard]

This requires a discussion as it is not meant to be used as a soft binding mechanism for manifest recovery.

[domguinard]

This entry was discussed during the WM TF meeting on Oct 28, as this algorithm is not meant to be used for the purpose of C2PA Manifest recovery it does not respect the following criteria:
The algorithm covered by the entry shall support embedding, extracting, or both, of a soft binding identifier for the purpose of manifest recovery.
and hence cannot be accepted.