ElevenLabs Music force_instrumental and seed vs Sume prompt-only

ElevenLabs' Compose endpoint has a force_instrumental boolean that guarantees an instrumental result and a seed integer for more consistent results. Sume's Music Router has no equivalent fields: instrumental is a sentence in the prompt, and every generation is a fresh take.

The ElevenLabs fields

On the Compose reference, force_instrumental defaults to false and, if true, guarantees the song is instrumental (read 2026-10-02). seed is optional and the page says the same seed with the same parameters can help achieve more consistent results (read 2026-10-02). There is also sign_with_c2pa, which applies only to mp3 files (read 2026-10-02).

What to do on Sume

End the prompt with one clause: Instrumental, no vocals. Add no spoken word only when narration will play over the track. The Music 1.0 docs say these are creative directions, not guaranteed output settings, and tell you to verify the generated audio.

Because a guarantee is missing, add a check step to your pipeline: listen or transcribe the file, and regenerate if vocals appear. Each accepted generation costs the fixed Music price, so cap retries.

Why there is no seed

Sume's Music 1.0 page states there is no seed, temperature, guidance or duration parameter. Google's docs for Lyria 3.5 say results may vary between calls even with the same prompt, which is the engine Music Router runs today.

If you need to reuse a take, keep the artifact. The completed job returns a media.sume.com audio URL, and that file is the only reliable way to get the same track again.