Gemini Omni video edit prompts that keep everything else the same

Keep the edit prompt short and end it with "Keep everything else the same." That is Google's advice for editing a video with Gemini Omni Flash: simple prompts work best, and overly descriptive prompts can lead to unintended changes.

The tips below come from Google's Gemini API page on Omni Flash (last updated 2026-09-23, read 2026-09-29). Sume lists gemini-omni-flash-1.1 for video edits on the Video Router, where you send video_url plus a prompt that describes the edit.

What does an Omni edit prompt look like?

Google gives four short examples, and the pattern is one change per prompt.

Why add "Keep everything else the same"?

Google says that when you edit one specific aspect of the video, including that sentence maintains visual consistency. Its own before and after pair: instead of a long sentence describing a black cat, its path across the screen and how he strokes it, write "Add a cat that jumps onto his lap, he begins to pet it. Keep everything else the same." For removing an object, Google's short form is "Make the phone invisible. Keep everything else the same."

What are the input limits for editing?

Google's limitations list says input videos for editing and extension must be 10 seconds or less when uploaded, and that voice editing is not supported. It also says editing or extending uploaded videos is not available in the European Economic Area, Switzerland and the United Kingdom, while editing videos the model generated is supported. Those are Google API rules; Sume's docs describe the edit request but do not list a per-region rule, so check your own account before you plan around it.

Can I edit a clip more than once?

On Google's Interactions API, yes: its guide describes stateful video editing, where you generate a video and edit it with follow-up prompts using a previous_interaction_id, and each turn builds on the previous result while preserving elements you did not mention. It also warns that setting store=false means the generated video will not be editable in later turns. That is a Google API feature. Sume's docs describe an edit request that takes a video_url and do not describe a conversation id. Keep each step to one change, as above, so a later edit does not undo an earlier one.

How do I send an edit on Sume?

Send video_url and the prompt to the Video Router. Sume's docs say resolution is optional for an edit (default 720p), that an edit takes no aspect_ratio or duration, and that video_url cannot be combined with image_url, end_image_url or reference_*_urls. Edit a video with a prompt walks through the full request.

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: omni-edit-prompt-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Change the lighting to be more dramatic. Keep everything else the same.",
    "video_url": "https://example.com/clip.mp4",
    "resolution": "720p",
    "mode": "async"
  }'