Macro liquid pour video prompt for Gemini Omni and its price

A macro pour shot works best as one continuous 5-second clip: say what is poured, into what, the camera distance, the light, and the sound of the liquid. With Gemini Omni on Sume that is about $0.63 at 720p and $0.19 at 360p for a draft. Ask for a single unbroken scene, as Google's Omni guide suggests (read 2026-10-07), or the model may cut away mid-pour.

Liquids are a test of physics, so budget two or three takes.

The prompt pattern

Use five slots in this order: subject, action, camera, light, audio. Keep each to a short phrase.

• Subject: 'amber honey in a glass dipper over a white ceramic bowl'.
• Action: 'honey pours in a thick slow ribbon and folds on itself'.
• Camera: 'macro close-up, locked-off, shallow depth of field'.
• Light: 'soft side light from a window, warm highlights'.
• Audio: 'faint viscous pouring sound, quiet room. No music. No dialogue.'

Request

Add 'single unbroken scene' to stop automatic cuts. The request below is a 5-second 360p draft; change resolution for the keeper.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: honey-pour-draft-1" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Single unbroken scene. Macro close-up, locked-off, shallow depth of field: amber honey pours from a glass dipper in a thick slow ribbon into a white ceramic bowl. Soft warm side light from a window. Audio: faint viscous pouring sound, quiet room. No music. No dialogue.",
    "duration": 5,
    "resolution": "360p",
    "aspect_ratio": "16:9"
  }'

Cost by resolution

Sume bills list times 1.25 per second and rounds each job up to the cent. fal's list for Omni Flash 1.1 (recorded 2026-08-28) is $0.03, $0.10, $0.15 and $0.30 per second for 360p, 720p, 1080p and 4K.

What to look for in the take

Check the stream: does it keep one thickness, or does it change viscosity halfway? Does the level in the bowl rise? Watch the dipper edge, because glass and transparent objects are where artifacts appear. Google notes that 1080p and 4K on Omni are upscaled outputs (read 2026-10-07), so a 4K re-run will not add real macro detail that the 720p take lacked; pay for 1080p at most unless you need the pixel count.

Vary one word per take, such as 'thick' to 'thin' or 'slow' to 'steady'. If you want the same pour over a specific bowl, send the bowl photo as a first frame; frame images are supported on this model.

Beyond the pour

The same slots work for other liquids: oil, milk, syrup and coffee all follow the subject, action, camera, light and audio pattern. Change the viscosity word first, because it controls how the model draws the stream. For a clean white background, say 'seamless white backdrop' and expect some shadow; for a dark premium look, say 'black glossy surface with soft reflections' and keep the light source low. Do not ask for a label or text on the bottle in a pour shot unless it comes from a first-frame photo, since small text on moving glass is the most likely thing to warp.