Lyria 3.5 blocks artist-voice prompts: how to write briefs that pass

Describe the sound, not the singer. The Gemini API changelog for September 3 says Lyria 3.5 is generally available with full-length songs, natural vocals and fine control over duration and structure, and its docs note that prompts requesting specific artist voices are blocked.

What the changelog says

Rewrite a blocked prompt

Replace a name with the qualities you hear in it: register, texture, pace, mood, and language. The aim is a brief a human producer could follow without knowing who you had in mind.

Steering structure on Sume

The Music Router takes a prompt of 1 to 5000 characters. Put exclusions in the positive prompt, because a non-empty negative_prompt is unsupported, and steer length and sections in the text, for example a 2-minute track or [0:00-0:30] Intro: …. duration and duration_seconds are rejected. With sume/music-auto the engine is Lyria 3.5 today, and job.request.routed_model tells you which one ran.

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: song-001" \
  -d '{
    "model": "sume/music-auto",
    "prompt": "A 2-minute track. [0:00-0:30] Intro: soft piano. Warm, breathy mezzo-soprano vocal, English lyrics about a long drive home. Instrumental break at 1:00."
  }'

If a prompt is refused

Remove every proper name of a performer, then run again. Keep the genre, tempo, instruments and mood. Read result.lyrics on the finished job, which carries the model-reported lyrics or section map when present, to see how the structure was interpreted.