Seedance 2.5 green-screen editing: swap a background on Sume

Sume does not expose a green-screen or editing mode for Seedance 2.5: the seedance-2.5 id takes text, frames and references, and a video input there is a reference, not an edit source. To swap the background of an existing clip on Sume, send the clip as video_url to gemini-omni-flash-1.1 with a prompt that describes the new background.

That is a different model, so the result will not match what Seedance's own editing would give. It does give you a hosted edit through the same API.

What the vendors say

ByteDance's Seedance 2.5 page mentions smarter reference and editing capabilities, with white-model control and green-screen editing. fal's page describes green-screen editing as swapping backgrounds while the subject reacts to the new surroundings. Neither page gives the request fields for it, and I found no Sume doc that exposes it, so I treat it as not available on Sume today.

The Sume route: Omni edit

Omni's edit mode is triggered by video_url. You cannot combine it with image_url, end_image_url or reference_*_urls. Native audio stays on, and the API rejects aspect_ratio in this mode because the source sets the output, and a duration is only a cost hint.

The edit lives on the Video Router surface. It is not part of the OpenRouter-shaped /v1/videos body.

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: bg-swap-001" \
  -d '{"model":"gemini-omni-flash-1.1","prompt":"Replace the background with a sunlit rooftop garden. Keep the person, their motion and the sound exactly the same.","video_url":"https://example.com/clip.mp4","resolution":"720p","mode":"async"}'

How the options compare

Use the table to choose by what you need to keep.

Know the limits

An edit regenerates the clip rather than compositing a cutout, so faces, hands and small details can shift. Check the result against the source before you ship it. There is no mask field, so you describe what to keep in the prompt.

If the background change is all you need and the subject has to stay pixel-identical, a model edit is the wrong tool; a matte-based workflow in your own editor will beat it.

Write the edit prompt well

Omni's edit is driven by one prompt, so the prompt has to say what stays as well as what changes.

• Name the thing to replace, in plain words: the wall, the street, the sky.
• State what must not move: the person, their motion, the camera path and the sound.
• Describe the light of the new background so the subject does not look pasted in.
• Keep it to one change per request. Two changes in one prompt more often shift something you wanted to keep.
• Check hands, hair edges and any text in the frame on the result before you ship it.