Gemini Omni prompt timecodes: [0-3s] and scene cuts

Gemini Omni Flash takes timing in plain words or a [0-3s] timecode syntax, so you can script cuts inside one clip. Google's examples and a 10-second Sume call.

4 min readSume
All posts

Yes: Gemini Omni Flash can be told when things happen. Google's prompt guide says you can prompt for events at specific times in plain language, such as "After 3 seconds, a woman enters the scene", or in a timecode syntax such as [0-3s] followed by the action. That lets one clip carry its own cuts and rhythm.

The syntax and examples below are quoted from Google's Gemini API page on Omni Flash (last updated 2026-09-23, read 2026-09-29). They are Google's prompting advice, not a Sume feature: Sume's Video Router docs list gemini-omni-flash-1.1 with a prompt field and 3–10 second clips, and do not describe timecode syntax themselves.

What timing syntax does Google show?

Google says there is no precise syntax needed for natural language, and shows both forms.

From Google's Omni Flash prompt guide, read 2026-09-29.
FormExample from Google
Natural languageAfter 3 seconds, a woman enters the scene.
Natural languageAt 5s the chorus starts in the background audio.
Cut rhythmEvery 2s cut to a new frame.
Timecode[0-3s] A person is walking, [3-6s] They stop and turn around, [6-10s] They start running

How do I write scene cuts inside one clip?

Google calls timing prompts especially useful for creating your own scene cuts, rhythm or rapid-fire sequences, and gives "In a rapid fire sequence, every half a second (12 frames at 24fps) change the scene to a new location". Because Omni Flash makes a few shots by default, Google's guide also says to prompt for a single continuous shot if you do not want cuts. Give each time slice one action, and end the last slice at the clip's length.

How long can the timeline be?

On Sume the model accepts 3–10 second clips, so a full timecode script fits in [0-10s]. Google's own six-reference-image example writes three slices, [0-3s], [3-6s] and [6-10s], so its last slice ends at 10 seconds, the top of that range. For extensions on Google's API, its guide adds that 0s refers to the beginning of the extended part; Sume's docs list no extend call, so that rule does not apply to Sume requests.

How do I send a timecoded prompt on Sume?

Put the whole timecoded prompt in prompt and set duration to at least the last timecode. Whether the model honors a given timecode is Google's model behavior; test a short clip first.

curl -X POST "https://api.sume.com/v1/videos" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: omni-timecode-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "[0-3s] A person is walking down a street. [3-6s] They stop and turn around. [6-9s] They start running.",
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "duration": 9
  }'

Can I mix timecodes with reference images?

Google's guide does: its six-reference example writes [0-3s], [3-6s] and [6-10s] slices that each name a reference with a tag such as <IMAGE_REF_0>. Sume's docs use the same tag style, counting from 0 in list order, for reference_image_urls (up to 10) and reference_video_urls (up to 3, each at most 3 seconds) on the Video Router. Gemini Omni Flash 1.1 API covers the reference modes.

Sources

Related posts

More in Models

All Models posts

Written by Sume