ElevenLabs Music v2.5 on Sume: 300 s max, started-minute billing

ElevenLabs Music v2.5 runs on Sume as elevenlabs-music-2.5, with duration_seconds from 10 to 300 (default 60). It is billed per started minute: 60 seconds is one minute, 61 seconds is two, and a full 300-second track is five.

fal, which serves the engine, lists "$0.6 per output audio minute" and says audio is "rounded up to the closest minute" (read 2026-10-14). Sume sells at the fal list price times 1.25 and shows credits only.

Request

Pin the id and the length together. Omit the length and you get 60 seconds.

As with the other length-controlled rows, image_url is refused, and so is duration.

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: eleven-120s-001" \
  -d '{
    "model": "elevenlabs-music-2.5",
    "prompt": "Cinematic strings build to a warm resolution, 90 BPM",
    "duration_seconds": 120
  }'

Minutes charged and fal list price

Because of the round-up, the cheapest long tracks end exactly on a minute boundary. A 120-second track and a 61-second track differ by one minute of billing; 120 and 121 differ by another.

When it is the right pick

This row is the one that reaches five minutes. The other length-controlled rows stop at 190 (Stable Audio 2.5), 240 (ACE-Step) and 180 seconds (CassetteAI). If you need a single long bed under a talking-head video, it avoids stitching.

If three minutes or less is enough, compare the other rows first: they cost less per minute on fal's list. The Music Router docs list every range, and the live credits are on the API rate card.