A 3-second sting from an Omni clip's native audio for $0.01

Omni clips always carry native audio. Audio detach with a 0 to 3 s range pulls that sound out as a wav for $0.01, against $0.125 for a new Lyria bed.

4 min readSume
All posts

You can take the first three seconds of the native audio of a Gemini Omni Flash 1.1 clip and keep it as a separate wav file for $0.01, using one Sume audio detach job with range: { start: 0, end: 3 }. Omni's audio is always on, so the clip you already paid for carries a sound you can reuse as a transition or end-tag sting.

This is cheaper than generating a sting ($0.125 per Music generation), but you do not control what the sound is. A clip prompted for a product shot may carry room tone, an impact, or speech. Listen before you use it.

The request

Audio detach takes one Sume-hosted video (video_url on media.sume.com), needs an Idempotency-Key header and returns a new audio artifact. The default format is sample-exact wav. The video is untouched.

{
  "video_url": "https://media.sume.com/artifacts/artf_demo/omni.mp4",
  "range": {
    "start": 0,
    "end": 3
  },
  "format": "wav"
}

Cost and limits

The job is worker ffmpeg with no provider inference, so the price is flat. The table compares routes to a 3-second sting.

Routes to a 3-second sting, Sume list prices as of 2026-10-08
RouteCostWhat you control
Detach 0 to 3 s of an Omni clip you already have$0.01Only the range, not the sound
Detach from a new 3-second Omni clip (360p)3 x $0.0375 + $0.01 = $0.1225The prompt, loosely
Generate with Music 1.0 or Music Router$0.125A brief with tempo, key and instruments
Generate, then cut to 3 s with a Timeline audio split$0.125 + $0.01 = $0.135The brief and the cut

Caveats

The source must have an audio track. A clip with none fails with detach_source_has_no_audio, but Omni clips always have one. The range must end after it starts (audio_detach_range_empty otherwise, and a range longer than 900 seconds is also refused), and the source can be at most 1800 seconds. A 3-second range is nowhere near these caps.

Two practical points. If you detach as mp3 (128 kbps), a 3-second file is about 48 KB; the default wav is sample-exact and the safer format for any later join, since the Timeline docs say mp3 adds priming padding at every edge. And the native audio of a clip is part of that clip: if you re-use it commercially, you are re-using what the model generated, so treat it like any other generated asset and check the clip as you would before publishing.

A short fade helps. A 3-second cut taken mid-sound can start or end on a click. Two ways to avoid it: choose a range that begins at a quiet moment, or lay the sting in a Timeline with a short fade. The Timeline docs allow output fades up to 5 seconds, and the soundtrack field takes fade_out_seconds up to 10, so either route is possible without another generation. If the sting is meant to be heard at the end of an ad, put it last on the spine and leave the bed's own fade-out to finish underneath it.

When a new bed is the better buy

A detached sting is a bargain when the clip's sound already fits. When it does not, $0.125 buys a brief you wrote yourself: a named tempo, key and a few instruments, with no vocals if the sting sits under a voiceover. Do not spend time listening to ten detached ranges to save $0.115; one Music generation is faster to judge.

Sources

Related posts

More in Media tools

All Media tools posts

Written by Sume