Explainer music that changes per scene: three Lyria briefs, one concat

For an explainer where each scene needs different music, make one Lyria 3.5 track per scene with POST /v1/music-router/generate, then join them with operation: "concat" on POST /v1/timeline-1.0/audio. The concat is gapless, sample-domain and costs $0.01 flat. Steer each track's length in its prompt, because duration and duration_seconds are rejected.

Source: Music Router and Timeline audio, read 2026-10-04.

Three briefs

Sume's music brief template says contrasting scenes should vary genre family, tempo by at least 12 BPM, and lead instrument. Three briefs below follow that.

Submit the first brief

The BPM gaps are 24 and 16, both above the 12 BPM rule. Put exclusions in the positive prompt, since a non-empty negative_prompt is rejected. Each request is one job, so use an Idempotency-Key per scene.

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: explainer-music-problem-001" \
  -d '{
    "model": "sume/music-auto",
    "prompt": "Tense minimal electronic, 78 BPM, A minor. Pulsing synth bass, sparse clicks, no drums until 0:20. A 30-second track. Instrumental, no vocals."
  }\

Join them

When all three tracks are ready, read each audio artifact from result.artifacts[]. Tracks must be on media.sume.com and share a channel layout, or concat fails with audio_parts_channel_mismatch. Concat takes 1 to 20 parts.

curl -X POST https://api.sume.com/v1/timeline-1.0/audio \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: explainer-music-concat-001" \
  -d '{
    "operation": "concat",
    "parts": [
      { "url": "https://media.sume.com/artifacts/artf_demo/problem.wav" },
      { "url": "https://media.sume.com/artifacts/artf_demo/solution.wav" },
      { "url": "https://media.sume.com/artifacts/artf_demo/cta.wav" }
    ]
  }\

Line up the picture

Lyria does not guarantee exactly 30 seconds, so read segments[] from the concat result. It gives each part's start and duration_seconds, which are the offsets to re-base Timeline 1.0 video[].start against. Use the merged file as audio.url on the render.