Gemini Omni C2PA credentials on Google's Agent Platform vs a Sume clip
Google says C2PA content credentials and SynthID are on by default for Omni Flash on its Agent Platform. What that means for a clip you generate on Sume.

Google's Cloud post says C2PA content credentials and SynthID watermarks are enabled by default for Gemini Omni Flash on its Gemini Enterprise Agent Platform. The Gemini API page and the Vids post mention SynthID only. Sume's docs do not say whether provider credentials survive into a clip you generate through Sume, so do not promise a customer that they do.
What does Google say, surface by surface?
The three pages I read do not say the same thing, and the gap matters if you must show provenance. Read the exact wording before you write a compliance answer.
Google's Cloud blog table lists C2PA content credentials as enabled by default and SynthID as imperceptible and enabled by default for the Agent Platform. Google's Gemini API Omni page says SynthID invisible watermarking is included and that policy violations are blocked. The Vids post says a SynthID digital watermark is embedded in the video frames.
| Google page | SynthID | C2PA |
|---|---|---|
| Cloud blog, Agent Platform | Enabled by default | Content credentials enabled by default |
| Gemini API Omni page | Included | Not mentioned on the page |
| Google Vids post | Embedded in the frames | Not mentioned on the page |
What does that mean for a Sume clip?
SynthID is described as embedded in the video frames, which is a property of the pixels, not a sidecar file. C2PA credentials are metadata attached to the file, and file operations can drop metadata. I could not find a Sume doc that says whether Omni's credentials are present on the artifact it returns, so the only honest check is to test the file you receive.
Sume's media tools return new MP4s. Video trim and the Timeline render both create new files and do not document carrying metadata across, so assume a trimmed or assembled file needs its own check.
How do I check a clip I generated?
Download the artifact from the job result and run a C2PA reader against it, such as the open-source c2patool, before and after any trim or timeline step. If the manifest is gone after a step, the step removed it. That tells you which step to avoid or to repeat on a re-sign flow of your own.
The point of the test is the file you publish, not the file Google generated. Keep a note of each step's tool and version; the other provenance posts on this blog, including the C2PA soft binding check, explain what survives re-encoding.
c2patool omni-clip.mp4
c2patool omni-clip-trimmed.mp4
# compare: does a manifest still print after the trim?What should my disclosure say?
Say what you know. "Generated with Gemini Omni Flash" is true for any clip you made with it. "Carries C2PA credentials" is only true for a file you tested.
If a client needs signed provenance on the final MP4, plan for a signing step that you own after your last edit.
Sources
Related posts
More in Comparisons
- Gemini Omni multi-turn edits: previous_interaction_id vs Sume jobs
Google's Omni API chains edits with previous_interaction_id. Sume has no such field: each edit is a new job on a video_url. How to chain by hand.
- Google Vids makes free Omni 1080p scenes: when to use an API
Google Vids added Gemini Omni 1.1 Flash on Sept 23: 1080p scenes, extend and clip duration, free with a Google account. Where Sume's API route fits instead.
- GPT Image 2.5 high vs GPT Image 2 high: price per image
At 1024x1024 high, fal lists GPT Image 2 at $0.211 and GPT Image 2.5 at $0.05268, about a quarter. On Sume that is $0.264 vs $0.066 at list x 1.25.
- GPT Image 2.5 vs Gemini 3.1 Flash Image: price per 1K image
fal lists GPT Image 2.5 at $0.05268 per 1024x1024 high image; Google lists Gemini 3.1 Flash Image at about $0.067 per 1K image.
Written by Sume