Edit a finished Omni clip with video_url instead of re-rolling it
Send an existing clip as video_url to gemini-omni-flash-1.1 on Sume with an edit prompt. Resolution defaults to 720p; send no duration or aspect_ratio.

To change one thing in a clip you already have, send it as video_url to gemini-omni-flash-1.1 with a prompt that states the edit. On Sume that request is routed as a video-to-video edit. resolution is optional and defaults to 720p, and you must not send aspect_ratio or duration, since the edit keeps the clip's own shape and length (Sume docs: Video Router, read 2026-10-05).
The request
The clip has to be a Sume-hosted artifact URL, such as the output of an earlier job. The prompt describes the change, not the whole scene again.
curl -X POST https://api.sume.com/v1/videos \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Idempotency-Key: edit-shot3-raincoat-v1" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-omni-flash-1.1",
"prompt": "Change the mustard raincoat to dark green. Keep the motion and the street unchanged.",
"video_url": "https://media.sume.com/artifacts/artf_demo/shot3.mp4"
}'Edit or re-roll
Google describes Omni 1.1 Flash as supporting video references and extension; it does not describe a price for edits (Google, read 2026-10-05). Sume bills the provider list times 1.25 for each output second at the chosen resolution, so an edit of a 10-second clip is billed like a 10-second output. Check the catalog and the pricing page for the live rate.
| Situation | Better choice | Why |
|---|---|---|
| One detail is wrong; the motion is right | Edit with video_url | Keeps the take you approved |
| The motion or framing is wrong | Re-roll with a new prompt | An edit cannot rebuild the shot |
| The first or last frame must change | New image-to-video job with a corrected still | Frames are controlled by the images |
| You need a different length | New generation | Edits take no duration |
Habits that keep edits cheap
If the edited clip is the one you will ship, finish it at the resolution you need as the last step, then join it with the other shots in a timeline.
One more rule: the edit needs a Sume-hosted clip, so if the footage came from elsewhere you first need it on Sume's media host. Generate-then-edit within the same workspace is the simple path, because every finished job returns a durable artifact URL you can pass straight on.
Keep both versions. The original clip is your fallback if the edit changes something you did not ask for, and the pair also tells you, in a side-by-side, whether the edit did what the prompt said. Delete nothing until the final is approved.
If the edit is part of a longer piece, do it before the timeline join, not after. The timeline takes finished clips and does not regenerate anything, so a fix made earlier flows into the render without extra work.
Write the prompt as an instruction with a boundary: say what changes and what must stay. Phrases such as keep the motion and the street unchanged give the model a reason to leave the rest alone. A vague prompt such as make it better invites changes you did not want, and you pay for each attempt.
- Edit the draft, not the final, when you are still testing the instruction. Send
resolution: 360pon the edit to keep the pass small. - Edit one thing per request. A prompt that changes colour, time of day and a prop at once is a coin flip.
- Read the result before you chain another edit. Edits stack their errors.
- Use an Idempotency-Key that names the shot and the edit, so a retry replays the same job and a changed body is refused with a 409.
Sources
Related posts
More in Developers
- Edit a photo with an Agent Completion: attach it, read output.images
Send the photo in attachments to POST /v1/agent/completions with a spend cap, poll the run, read the edit from output.images. Python code and limits.
- Per-request character limits: Eleven v4 10,000 vs Sume TTS 20,000
Eleven v4 takes up to 10,000 characters per generation, Sume TTS 1 to 20,000. A 45,000-character script is 5 requests versus 3, with a chunking script.
- ElevenLabs Music can sign MP3s with C2PA; what to log for a Sume track
The ElevenLabs Music API has sign_with_c2pa for MP3 output only. Sume's Music docs name no audio credential, so keep a record of job, prompt and routed model.
- Edit a transcript with plain-language instructions: ElevenLabs STT
ElevenLabs STT edits a transcript from an instruction of up to 2,000 characters and returns edited_transcript. Sume captions align your own script_text.
Written by Sume