ASMR-style AI video prompt: Gemini Omni with sound, 10 seconds

Gemini Omni can make an ASMR-style clip because its audio is generated with the picture, not added later. Prompt for a close shot, slow repeated hand motion, and name each quiet sound; on Sume a 10-second 720p take costs $1.25. Sound is always on for this model, which suits this genre, but the quality of tiny sounds is something you can only judge by listening to headphones, so draft first.

This is a style prompt, not a technical guarantee. The model may smooth or skip the faintest sounds.

What ASMR prompts need

ASMR depends on three things: proximity, repetition and clean quiet. Write all three.

• Proximity: 'extreme close-up, hands only, soft focus background'.
• Repetition: 'slow rhythmic tapping on a wooden box, same pace throughout'.
• Clean quiet: 'room is silent except for the tapping and fabric rustle. No music. No dialogue.'
• Camera: 'locked-off, no cuts', which echoes Google's 'continuous shot' advice (read 2026-10-07).

Request

Ten seconds is the Omni maximum, and long takes help rhythmic content. The prompt below stays within it.

import os, requests

prompt = ('Extreme close-up, locked-off, single continuous shot. Two hands tap slowly '
          'on a wooden jewelry box, then brush a velvet cloth across the lid. '
          'Soft diffused light. Audio: crisp tapping, gentle fabric rustle, '
          'near-silent room. No music. No dialogue.')
r = requests.post(
    'https://api.sume.com/v1/videos',
    headers={'Authorization': 'Bearer ' + os.environ['SUME_API_KEY'],
             'Idempotency-Key': 'asmr-box-1'},
    json={'model': 'gemini-omni-flash-1.1', 'prompt': prompt, 'duration': 10,
          'resolution': '720p', 'aspect_ratio': '9:16'},
    timeout=60)
print(r.status_code, r.json())

Cost

List rates are $0.03, $0.10, $0.15 and $0.30 per second at 360p, 720p, 1080p and 4K in the Video Router doc; Sume bills list times 1.25, rounded up to the cent per job.

Limits to plan around

There is no audio input on this model, so you cannot supply your own recorded sounds; the sound is whatever Omni generates. Because there is no seed on Sume, the same prompt gives a different take each time. If you need cleaner sound, the audio detach endpoint can pull the track out of a finished clip for further work, at a documented $0.01 per job. Keep the clips short and loop-friendly if you will repeat them in a longer edit.

Making the result useful

ASMR clips tend to be watched on headphones and looped. Check the first and last half-second for a click or a jump, because those are the points a viewer hears on repeat. If you need a longer piece, generate two or three different 10-second takes and place them in sequence in a timeline rather than asking for a longer clip, which Omni does not allow. Keep the visual simple; a busy scene competes with the sound. If the audio is good but the picture is not, you can detach the track and reuse it, within the rights you have to the generated material.