AI song structure tags: [Verse], [Chorus] and timestamps

Structure an AI song by putting section tags or timestamps in the prompt. Google's Lyria page says to use tags like [Verse], [Chorus] and [Bridge], and also documents timestamp sections such as [0:00 - 0:10] Intro: .... Sume's docs show the same timestamp style, [0:00-0:30] Intro: ..., and send the text to Lyria 3.5 through the Music Router.

Google's facts are from its Lyria page; Sume's from Music 1.0 and the Music Router. Read 2026-09-29.

Which structure syntax works?

What does a structured request look like?

The tags are plain text inside prompt; there is no separate structure field. Combine them with tempo, key and instruments.

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: structure-001" \
  -d '{
    "model": "lyria-3.5",
    "prompt": "Anthemic pop rock, 124 BPM, E major, no vocals. [0:00-0:15] Intro: clean guitar. [0:15-0:40] Verse: bass and drums enter. [0:40-1:00] Chorus: full band, big drums. A 1-minute track."
  }'

Do the sections have to land exactly?

No guarantee. Sume's docs describe prompt directions as creative guidance and tell you to verify the audio; Google notes results can vary between calls. Listen for where the chorus actually starts, and adjust the times in a new take.

Where do I limit the prompt?

A prompt is 1 to 5,000 characters. Structure a song in a few tagged lines rather than a long essay, and put exclusions in the positive text, since a non-empty negative_prompt returns HTTP 400.