Does an AI video API add a C2PA record or watermark to output?
Sume's docs do not say its video output carries a C2PA record or watermark, and trim and Timeline re-encode files. How to check a file and what to do instead.
Sume's documentation does not say that its video output carries a C2PA record or a watermark, so do not assume either. Whether a mark exists depends on the model behind the job, and the Sume docs pages do not describe it for any of them. Test the file you receive and disclose the video yourself.
The Sume facts below come from the video generation, video trim and video inspect docs. The YouTube facts come from YouTube Help, read 2026-09-29.
What do Sume's docs actually say about output?
They say what the file is and where it lives, not how it is marked.
| Question | What the docs state |
|---|---|
| Where does a trimmed clip go? | A new MP4 artifact; the source is untouched |
| How is the default trim cut? | precision: exact is a frame-accurate re-encode (libx264, yuv420p) |
| What does the inspect probe report? | Duration, size, rotation, fps, codecs, audio and HDR |
| Is a watermark or C2PA record described? | Not on these pages |
Does re-encoding remove a C2PA record?
Sume's docs do not say. Trim's default mode re-encodes the video, and Timeline returns one newly rendered MP4, so a record attached to a source file cannot be assumed to be present on the new one. If your workflow depends on a record, run your own C2PA-capable reader on the final file, not on the clip you started with.
Why does the answer matter for YouTube?
YouTube says it may automatically apply an AI label to content made with its own AI tools or that carries C2PA metadata. That is a bonus, not a substitute: the same page says creators disclose realistic content that makes a real person appear to say or do something they did not, alters footage of a real event or place, or generates a realistic scene that did not occur. See YouTube Help.
What should I do instead of relying on a mark?
Put the disclosure in your own publish step. A short on-screen line or caption and the platform's own AI toggle work whether or not the file carries a mark.
- Run your own reader on the final MP4, not the source clip.
- Note the model id of each clip in your project log.
- If a client needs a provenance record, ask the model vendor's own page, not Sume's docs, what it adds.
How do I read the file Sume gives back?
A finished trim job returns a video_url for a new artifact, never the source. Download that final file and run your C2PA-capable reader on it, then on your published upload, since a platform may re-encode again. The docs do not promise either file carries a record, so the check is the only way to know.
Sources
Related posts
More in Use cases
- Does TikTok Shop allow slideshow videos? Not on their own
TikTok Shop's Seller Center page dated 09/17/2026 bars still, looping or scrolling images through a whole posted video. What that means for AI video.
- Do I need to label AI voiceover on YouTube Shorts?
Sometimes. YouTube requires a label when a synthetic voice makes a real person seem to say something, and lists AI music as an example. Own-voice clones do not.
- Donor thank you video: one render, each donor's name
A donor thank you video can show each donor's name for the price of one render plus a caption job per name, or say each name aloud at a full render per donor.
- Translate a video to another language with AI voice, on time
A translated voice rarely lasts as long as the original. Use TTS sentence segments, the speed control and Timeline audio offsets to keep each line on its slot.
Written by Sume