Music 1.0 to Music Router: what changes in your integration

Music 1.0 is being retired gradually on Sume, and its routes keep working. Every request already resolves through the Music Router (sume/music-auto), so the migration is small: for new code, POST to /v1/music-router/generate and optionally send a model id. The body, the price ($0.125) and the job flow stay the same.

What stays the same

The Music Router body is the same as for Music 1.0, plus an optional model field. The prompt is required, 1 to 5,000 characters. image_url is optional, a public HTTPS image, or null to clear it. A non-empty negative_prompt is not supported: omit it or send an empty string. Sume rejects duration and duration_seconds, so you steer length in the prompt.

Old routes keep working and keep job.model set to sume/music-1.0, so existing dashboards that group by model do not change under you.

What changes

First, you can pin an engine. If you omit model or send sume/music-auto, Sume picks the engine (Lyria 3.5 today). Unknown ids fail with 400 model_not_found and a catalog_url. Second, job.model echoes the id you requested, and job.request.routed_model names the engine that actually ran, on both the router and the Music 1.0 routes. Log that field, because auto can move.

Third, the model-run alias for Music 1.0 (POST /v1/models/sume/music-1.0/runs) is the legacy path; new work should use the router URL.

A safe cutover

Change the URL and nothing else, and run a few generations. Compare the audio artifact in result.artifacts (type audio) and read result.lyrics when it is present; it is model-reported metadata, not a measurement of the audio. Keep your Idempotency-Key scheme the same so a retry during the cutover cannot create a second charge.

Once the router path is stable, decide whether to pin lyria-3.5 or leave auto. The related post on pinning covers that choice.

Checklist

Switch the URL. Log job.request.routed_model. Keep Idempotency-Key per request. Do not send duration, duration_seconds, or a non-empty negative_prompt. Decide whether to pin an engine. Test with one generation ($0.125) before you move traffic.

Nothing in the docs requires you to move on a deadline; the retirement is described as gradual, with old routes continuing to work. The recommendation is for new integrations.