Slow-motion water splash prompt for Gemini Omni: what 'slow' does

A slow-motion splash is a prompt about time, not a setting: Gemini Omni has no frame-rate or speed parameter on Sume, so you describe the event as happening slowly and in detail. Say 'ultra slow motion, droplets frozen mid-air, crown splash forming' and keep the clip to 5 seconds, which costs $0.63 at 720p.

The model decides how slow slow is. Judge the result by watching one droplet, not the whole frame.

Describe the physics you want to see

Models handle a named event better than a generic 'slow motion'. A fruit hitting water has a recognisable shape: a crown, rising droplets, a rebound.

• Event: 'a ripe lemon drops into clear water'.
• Detail: 'a crown-shaped splash forms, droplets hang in the air, bubbles stream upward'.
• Speed: 'ultra slow motion, every droplet visible'.
• Camera: 'locked-off, side view through glass, macro depth of field'.
• Sound: 'a deep soft plunk, then muffled bubbling. No music. No dialogue.'

Request and fallback

Ask for one splash only; a second drop in the same shot doubles the physics to get right. The request is a 5-second 360p draft.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: lemon-splash-draft-1" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Ultra slow motion, locked-off side view through glass, macro depth of field. A ripe lemon drops into clear water, a crown-shaped splash forms, droplets hang in the air, bubbles stream upward. Single continuous shot. Audio: deep soft plunk, muffled bubbling. No music. No dialogue.",
    "duration": 5,
    "resolution": "360p",
    "aspect_ratio": "16:9"
  }'

Cost

Sume bills list times 1.25 per second and rounds each job up to the cent. The Video Router doc lists $0.03, $0.10, $0.15 and $0.30 per second.

If the clip is not slow enough

If the splash plays at normal speed, do not raise the resolution; rewrite the speed phrase and shorten the event. A second route exists: trim the best second of the take with Sume's video trim endpoint and stretch it in your editor. That is a speed change of real frames, not new detail, so droplets will look smoother only if the source frames are clean. Remember that Google lists 1080p and 4K on Omni as upscales (read 2026-10-07); fine droplet detail comes from the prompt and the take, not from the resolution setting.

Where to spend the money

Draft the motion at 360p and only upgrade the take that has the best splash. Higher resolution will not repair a wrong shape. If you need a precise frame, such as the instant the crown peaks, pull stills from the finished clip with the video frames endpoint, which extracts frames at times you name. Use the best still in a thumbnail or a poster and keep the clip for motion. Sume jobs do not take a seed, so never expect a re-run to reproduce a lucky splash.