Edit a 10-second product clip with Omni Flash 1.1: $1.50 list, $1.88

Editing a 10-second product clip with Gemini Omni Flash 1.1 at 1080p lists at $0.15 per second on fal, so $1.50. Sume reserves the provider list price times 1.25 at submit, which makes it $1.875 for the same clip, about $1.88. The source clip sets the output length, so a longer source costs more.

List price and Sume's multiple

The vendor page for the edit endpoint, read 2026-10-08, lists per-second prices by resolution and gives the example that a 10-second 1080p clip costs $1.50 and the same clip at 360p costs $0.30. Sume's video docs state that at submit Sume reserves provider list x 1.25 for each model.

How the edit call differs

Edit mode is a video_url on the Video Router. The model has one catalog id, and the input shape selects the mode. Send prompt, video_url, and resolution. The source clip sets the output, so the edit sends no aspect_ratio (a 400 if given), and duration is only a reserve hint with a default of 8 seconds.

Do not send generate_audio: false, bitrate_mode, or reference_audio_urls. Each is a 400, because the model always produces native synced audio and has no audio reference. You also cannot combine video_url with image_url, end_image_url, or reference_*_urls. To condition a new generation on a clip, use reference_video_urls instead.

{
  "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": "720p"
}

Edit or regenerate

An edit keeps the motion and framing of what you shot. A regeneration invents new ones. If your problem is one object in a good take, such as the wrong flavor on a bottle, edit. If the take itself is wrong, regenerate. At 720p the edit of a 10-second take is $1.25 on Sume, which is cheap against a reshoot.

Always compare the result against the source before you publish. Pull matching frames at the same times from both clips and check the parts you did not want changed.

A worked comparison

Say the ad is a 10-second product take and the cup in it is the wrong color. A reshoot has crew time. A fresh generation gives you a different clip. The edit, at $1.875 for 10 seconds at 1080p, keeps the camera move, so for a one-object fix it is the better trade.

The arithmetic is simple: seconds x list rate x 1.25. For 6 seconds at 720p that is 6 x $0.10 = $0.60 list, $0.75 on Sume.

Budget

Because the reserve is a hint when you pass duration, set it close to the true length of the source. A clip that is 10 seconds with the default 8-second hint reserves for 8 and the output follows the source. Reading the receipt after the job tells you what was captured.