AI birthday song generator by API: Lyria 3.5 on Sume

You can generate a birthday song by API with one call to the Sume Music Router, which runs Lyria 3.5. Google's Gemini post lists custom birthday tracks as one of the templates for Lyria 3.5 in the Gemini app, and the same model is routable on Sume at a fixed $0.125 per generation. Google facts read 2026-10-01 from its Gemini blog; Sume facts from the Music Router and Music 1.0 docs.

How do I ask for a birthday song?

Put the name, the age or milestone, the genre and whether it is sung in the prompt. The prompt takes 1-5000 characters. Section tags such as [Verse] and [Chorus] steer structure, and a short timestamp marker like [0:00-0:30] steers length, since there is no duration field.

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: bday-maya-001" \
  -d '{
    "model": "lyria-3.5",
    "prompt": "Upbeat acoustic pop, 112 BPM, G major, handclaps and ukulele. Warm female vocal. [Verse] Happy birthday to Maya, thirty and bright [Chorus] Sing it loud tonight. About 1 minute."
  }'

What do I get back?

A job. Poll it or use a callback, then read the result: an audio artifact on media.sume.com (audio/mpeg) and, when the model reports them, the lyrics in result.lyrics. The job request shows which engine ran under routed_model.

Can I personalise at scale?

Yes, one job per recipient. Jobs queue when your workspace is at its concurrency limit, so a list of 50 names is accepted and works through. Use a different Idempotency-Key per recipient so a retry never makes a second song.

What should I not promise?

Do not promise exact length or an exact tune. The docs treat prompt directions as guidance, and there is no seed, so the same prompt can sound different each time. Listen to the track before you send it.