Seedance 2.5 C2PA, watermark and real-face rules: an ad checklist

The BytePlus Seedance 2.5 page lists three safety features: C2PA Content Credentials, visible watermarks, and restrictions on real faces with authorized exceptions handled through the ModelArk console. For an ad team that means three things to confirm per clip, and one thing to record.

What the page lists

Each feature is a vendor statement about the direct BytePlus path. None of them proves anything about the file you end up with after your own edits, or about a different platform serving the same model.

Before generating

Start with the brief, before any generation. If a person in the script is a real individual, such as a founder, creator, or celebrity, assume the real-face restriction applies and find the authorization route before you pay for generations.

• List every on-screen person and mark each as fictional or real.
• For real faces, find the authorization path for the exact access route you will use.
• Decide whether the delivery channel accepts a visible watermark.
• Decide who owns the disclosure label for the finished ad.

After generating

After the job completes, download the file and test it, rather than trusting the feature list. A credentials checker run on the delivered file answers the C2PA question directly. Edits that re-encode the video can change what survives, so test the final cut as well as the raw clip. The stored post on checking the delivered file walks through that test.

Roles: who signs off on what

Safety features cut across teams, so name an owner for each before production starts. A creative lead usually decides the brief and whether any real person appears. A producer owns the delivery spec, including watermark handling. Legal or compliance owns disclosure language and the real-face authorization record.

Without named owners, the usual failure is that everyone assumes someone else checked the credentials on the final export. A single sign-off row per clip, with a name and a date in each cell, prevents that. It also gives you something to show a client who asks months later how a clip was made.

• Creative: brief, people on screen, real or fictional.
• Production: watermark policy, final export spec, credentials check.
• Compliance: authorization record, disclosure label, retention of records.

What to record

Sume mirrors generated outputs into Sume-owned media URLs and tells integrations to store the Sume URL, not raw provider URLs, per the media inputs docs. Keep the job id with the clip, along with the model id from the poll response and the final cost in usage.cost. The video generation docs show those fields.

A per-clip row with job id, model id, prompt, reference inputs, and the answer to each question above is enough for most client sign-offs. The related post on per-clip disclosure records lists more fields.