Lyria 3 Clip is fixed at 30 seconds: which Sume music id to use

Lyria 3 Clip is fixed at 30 seconds. Sume's Music Router does not list a clip id: its ids are sume/music-auto (the default, which resolves to lyria-3.5), lyria-3.5 and lyria-3-pro. For a 30-second bed, send sume/music-auto and ask for "a 30-second track" in the prompt. Each generation costs $0.125.

What Google says

The Gemini API music generation guide describes two models. Lyria 3 Clip (lyria-3-clip-preview) is fixed at 30 seconds, MP3. Lyria 3.5 (lyria-3.5) makes full songs of about a couple of minutes, controllable with the prompt, MP3 by default with WAV optional. Both are 44.1 kHz stereo and accept text plus up to 10 images. All audio carries a SynthID watermark, and results are non-deterministic.

What Sume accepts

Sume does not take a duration or duration_seconds field on music; the API rejects both. An optional single image_url can set the mood.

How to get a 30-second track

Put the length in the brief and name where the track changes. A 30-second bed has room for an intro, one lift, and an ending, so write it that way.

curl -X POST https://api.sume.com/v1/music-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: bed-30s-001" \
  -d '{
    "model": "sume/music-auto",
    "prompt": "Bright acoustic pop, 100 BPM, G major. Fingerpicked guitar, soft claps, a glockenspiel lift at 0:15. A 30-second track. Instrumental, no vocals."
  }'

Check the length you got

Because length is steered by words, it is not exact. Read the duration from the artifact and, if the track runs long, trim it in Timeline with the soundtrack fade_out_seconds (up to 10) rather than regenerating. See exact length without a duration field for the trim route.

Because results are non-deterministic, two runs of the same prompt differ. If you need a choice, run a few and audition them.

Picking between the three ids

Leave the model as sume/music-auto unless you have a reason. It resolves to lyria-3.5 today, and job.request.routed_model tells you which engine ran. Choose lyria-3-pro when you specifically want that engine; the comparison post walks through the choice.

Whatever you pick, the price per generation is the fixed $0.125, so the choice is about sound, not cost.