Sonilo segment-level music controls vs Sume section markers

Sonilo's June 22, 2026 release says its text-to-music model, also on fal.ai, has segment-level controls that let you define different musical styles, moods and structures across sections of a track. Sume has no separate per-segment field. You describe the sections inside one prompt, using timestamped markers such as [0:00-0:30] Intro: ..., which the Music 1.0 docs give as the way to steer structure and length. The Sonilo statement is from its press release; the release does not list the field names.

How do section markers work on Sume?

They are text in the prompt, not parameters. The Music Router docs say to steer length in the prompt ("a 2-minute track" or markers like [0:00-0:30] Intro: ...), and that exclusions go in the positive prompt because a non-empty negative_prompt returns 400. The prompt can be 1 to 5000 characters, so a few sections fit comfortably.

What does a sectioned prompt look like?

Keep each section to one mood and one instrument change, and name one arc moment, so the model has something concrete. The Music 1.0 docs also suggest a tempo as a number and a key.

[0:00-0:10] Intro: sparse Rhodes, hushed, 84 BPM, D minor.
[0:10-0:25] Build: brushed drums enter, muted trumpet answers the Rhodes.
[0:25-0:30] Resolve: drums drop out, one held chord.
A 30-second track. Instrumental, no vocals.

What if the sections come out wrong?

Sume's docs call prompt directions creative guidance and tell you to verify the generated audio. There is no seed, so rerunning the same prompt can give a different result. Revise the section text and run a new job; the stored guide AI song structure tags covers the tag styles that Lyria accepts.