Omni Flash edit or regenerate: when video_url beats a new prompt

Edit with video_url when the shot is right and one thing is wrong: an object, a colour, a background detail. Regenerate with a new prompt when the motion, framing or pacing is wrong, because an edit keeps the source clip's timing and framing.

On Sume the edit runs on gemini-omni-flash-1.1 through the Video Router. A 6-second edit costs $0.75 at 720p and $1.13 at 1080p.

What an edit does

The edit request sends a prompt and the source video_url. The source clip sets the output, so the API sends no aspect_ratio and no duration to the provider, and a 400 comes back if you give an aspect ratio. If you send a duration with video_url, it is only the reserve-estimate hint, which defaults to 8 seconds.

Google's Gemini API video page describes Omni Flash as a model for video generation and conversational video editing with multi-turn edits (read 2026-10-07). Sume does not expose the multi-turn or extend interaction id; each edit is a single job.

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "Change the red jacket to dark green. Keep everything else the same.",
  "video_url": "https://example.com/street.mp4",
  "resolution": "720p"
}

Edit or regenerate

Use the table to decide before you spend. Both paths are billed at the same per-second rate for the resolution you choose.

Cost of a try

Because there is no seed, every run is a new take, edit or not. Budget for two or three tries. Prices are provider list per second times 1.25, rounded up to whole cents, on the length of the source clip.

Limits

The edit path always produces synced audio, so generate_audio: false and bitrate_mode are 400s, and reference_audio_urls is also rejected. The source clip cannot be combined with image_url, end_image_url or reference_*_urls. To use a clip as conditioning for a new generation instead, send it in reference_video_urls, which accepts up to three clips of at most three seconds each on this model.

• Edit is a Video Router field, not part of the /v1/videos body.
• Default edit resolution is 720p; ask for 1080p when you need it.
• Trim long sources first; the output length follows the source.