Does SynthID survive captions and a 9:16 crop on Omni clips?
Google says SynthID is built to survive cropping, filters and compression. What that means for a captioned or reframed Omni clip, and what Sume claims.

Google says SynthID is designed to survive common edits such as cropping, filters, frame-rate changes and lossy compression, so a captioned or reframed Omni clip should usually still carry it. Google's SynthID page, read 2026-10-08, does not promise detection after every edit and does not name captions or 9:16 reframes specifically. Treat the claim as a design goal, not a guarantee.
Sume's documentation makes no statement about SynthID. What Sume returns for an Omni job is the file from the provider, and what it does to files afterward is described below.
What Google says
Google's Omni page states that all outputs carry a SynthID watermark. Google's SynthID page describes it as an invisible watermark designed to withstand cropping, filters, frame-rate changes and lossy compression. Detection covers Google and partner content through the Gemini app, Search, Chrome and the SynthID Detector portal.
| Question | Google's page says |
|---|---|
| Is it visible? | No, invisible |
| Does Omni add it? | Yes, to all outputs |
| Edits it is designed to survive | Cropping, filters, frame-rate changes, lossy compression |
| Where it can be checked | Gemini app, Search, Chrome, SynthID Detector portal |
| Who it covers | Google and partner content |
What Sume does to the file
Sume's video trim and video filter endpoints are documented as server-side steps that return a new MP4. Trim cuts a [start, end) range and filter applies an ffmpeg operation. Both re-encode, which is a lossy-compression edit in Google's terms. Sume does not state whether a mark survives them, and does not check for one.
If provenance matters to you, test it on your own clip: run an Omni clip through your edit chain, then use one of Google's detection surfaces. Record the result with the date.
A practical rule
Keep the unedited provider file next to the edited one. If a platform or a client asks where the clip came from, the original is the better evidence. A caption added in your editor is an edit of the same kind as a crop. The watermark is a provenance signal for detectors; it is not a visible label for viewers, and it does not replace the disclosure that a platform may ask for.
How to test it yourself
A test takes ten minutes. Generate one 3-second Omni clip at 360p, which costs $0.1125 on Sume. Keep the original. Make three copies: one with a caption burned in, one cropped from 16:9 to 9:16, and one re-exported at a lower bitrate. Then open each copy in a Google detection surface listed on the SynthID page and note the result and the date.
Do the test again when your tools change. Google's page describes design intent for the watermark, and neither Google nor Sume promises a result for your particular encoder settings. If a copy fails detection, you have learned that for your chain, which is worth more than the general claim.
Sources
Related posts
More in Models
- Five requests Omni Flash cannot take, and what to send instead
Gemini Omni Flash 1.1 on Sume refuses clips over 10 seconds, silent clips, audio references, 1:1 and 2K or 480p output. What to send instead for each.
- Gemini Omni Flash 1.1 on Sume: spec sheet and billing in one page
Everything that ships for gemini-omni-flash-1.1 on Sume: 3 to 10 seconds, four resolutions, two ratios, always-on audio, reference limits, edit mode, billing.
- Gemini Omni rate limit: Google lists none, Sume lists them
Google's rate-limit page has no Omni or Veo row. It lists per-project RPM, TPM and RPD plus spend caps. Sume publishes per-key limits and concurrency by plan.
- GLM-5.3 reasoning cannot be turned off: cap the Sume run instead
Z.ai says GLM-5.3 always reasons, with low, high and max levels. What that means for run time and for a spend cap on a Sume agent run.
Written by Sume