Music prompt limit: 5000 characters is plenty for a Lyria brief

The Sume Music prompt must be 1 to 5000 characters. That is far more than you need: the docs' own example brief, covering mood, genre, tempo, key, instruments, arc and era, is about 330 characters. The limit matters only if you paste a long script or a list of sections, in which case trim before you send, since a longer request is rejected.

What a long prompt looks like

Section markers add length. Each marker such as [0:00-0:30] Intro: soft pad and muted bass costs about 40 to 60 characters. Eight markers plus a brief header stays near 600 characters. So the practical ceiling is not the count but how clearly the model can follow the plan.

Rules that interact with length

Exclusions go in the positive prompt, for example "no vocals, no spoken word". Do not send a non-empty negative_prompt; Sume returns 400 with negative_prompt_unsupported. Do not send duration fields. Put the length in the prompt instead.

Provider lyrics in the result are model-reported tempo and structure metadata. They are not an audio measurement, so they cannot tell you whether a long prompt was followed.

When shorter wins

A shorter prompt leaves less to contradict. If two instructions fight (for example 72 BPM and "driving, fast"), shorten and keep the number. The docs advise writing tempo as a number, key and mode by name, and one named arc moment, then ending with a single instrumental clause.

Each accepted generation costs $0.125 no matter how long the prompt is, so brevity helps quality, not price.

Sending it

The call is an ordinary POST to /v1/music-router/generate (or the Music 1.0 route). Poll the job, then read the audio artifact. The result is mostly audio/mpeg on media.sume.com.

• Async is the default mode.
• wait_timeout_seconds accepts 0 to 30 for sync or subscribe.