Swap a product's colorway in a finished clip: Gemini Omni edit

To make a colorway variant of a clip you already approved, use the edit capability of gemini-omni-flash-1.1 in the Video Router. Send the source as video_url and a prompt that gives the instruction, for example to change the bottle to another color and keep everything else. The docs example swaps a bottle for an apple with Replace the bottle with an apple. Keep everything else the same. The rules that decide whether your request is valid are on the Video Router page, and they are strict about what the edit cannot carry.

What the edit request accepts

video_url is the edit source, not a reference. You cannot send it together with image_url, end_image_url or any reference_*_urls. The edit takes no aspect_ratio and no duration, and resolution is optional with a default of 720p. Native audio is always on, and the API rejects generate_audio: false. There is no bitrate_mode and no reference_audio_urls. Sume bills the provider list times 1.25 for each output second, by resolution.

A colorway request

Name the thing to change, the new value, and what must stay. A short, concrete instruction is better than a description of the whole scene. One request makes one variant, so send one request per colorway, each with its own key.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: kettle-ad-colorway-sage-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Change the kettle from stainless steel to matte sage green. Keep the hand, the counter, the camera move and the audio the same.",
    "video_url": "https://example.com/approved-ad.mp4",
    "resolution": "720p",
    "mode": "async"
  }'

Limits to respect

Google describes video references of up to 3 seconds for this model, and the 360p draft mode as up to 60% faster and a third of the cost of standard 720p (read 2026-10-05). The Sume page does not state a source-length limit for the edit, and it lists 3 to 10 seconds for generation, so use a source of ten seconds or less until you have tested a longer one. If your ad is longer, edit it in segments and join them in Timeline.

The same idempotency rules hold for a draft and a final: the same key with a different body is a conflict, so a 360p test and a 720p final need two keys.

A batch plan

Take one approved clip, and list the colorways. Send one request per colorway, each with a key built from the clip id and the colorway name. Look at the first result before you send the rest: the risk is that the color change leaks into the hand, the label or the lighting. If it does, tighten the prompt with what must stay, and re-run the first one only.

Check the label text in every variant. A colorway change near a printed name can alter letters, and a retry is a new billed job. Keep a record of the source clip URL, the prompt text and the key for every variant, so a client who asks for the sage version again gets the same prompt and not a rewritten one.

Two limits of the approach are worth a test before you promise it. First, a color change on a reflective surface also changes what the surface reflects, so check the highlights and the hand. Second, the model is a general editor, not a product-recolor tool, so it can alter things you did not mention. The prompt line that says what to keep is the control you have, so write it every time, and keep it the same across variants so the differences you see come from the color.