Omni Flash edit: test at 360p, then run the final at 1080p

Run a Gemini Omni Flash edit at 360p first, then repeat it at 1080p once the prompt works. On Sume the rates are $0.0375 a second at 360p and $0.1875 at 1080p, so for an 8-second clip the draft is $0.30 and the final is $1.50. Each run is a different request, so give each its own Idempotency-Key; reusing a key with a changed body is a 409.

Two requests, one prompt

Edit mode is the video_url input on POST /v1/video-router/generate. The edit sends no aspect_ratio, and no duration reaches the provider: the source clip sets the output. A duration you send is only a hint for the reserve estimate, which defaults to 8 seconds. The docs say the reserve is a hint and not a probe, so the final charge is the thing to check on your job record.

The request

Keep everything but the resolution and the key the same, so that the draft says something about the final:

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "Replace the bottle with an apple. Keep everything else the same.",
  "video_url": "https://example.com/clip.mp4",
  "resolution": "360p"
}

Judging the draft

When the draft looks right, pull the first and last frame of both the source and the result, as in the frame-pair check, to see what stayed. Then send the same body with 1080p and the second key. If the draft fails because the model changed something you wanted to keep, tighten the prompt and re-draft; the 360p run is the cheap place to do that.