AI music brief template: seven axes for a Lyria 3.5 prompt

Write the music prompt as a seven-part brief: emotion, genre or lineage, tempo as a number, key and mode, two to four instruments with texture, an arc with one named moment, and era or production. Sume's Music docs recommend exactly these axes, because the music endpoint has no seed, temperature, guidance or duration parameter, and the prompt is the only control you have.

Google's Gemini API changelog (read 2026-10-02) says Lyria 3.5 went GA on 2026-09-03 with full-length song generation, text and image inputs, 44.1 kHz stereo, and duration and structure control. On Sume, you steer all of that in words.

What are the seven axes?

The table is the template from Sume's docs, with one example per axis. They are creative directions, not guaranteed output settings; verify the generated audio.

What does a filled brief look like?

Close with one clause, such as Instrumental, no vocals. Add no spoken word only when a narration will sit on top. The docs' own example reads: Hushed and slightly melancholic neo-soul nocturne, 72 BPM, D minor. Rhodes through tape wow, soft sub bass, brushed snare with rimshots, a single muted trumpet line. Sparse first half; the trumpet answers the Rhodes from 0:12 and the bass thickens for the last pass. Late-night, dry and close, 1998 production. Instrumental, no vocals.

Which fields should I not send?

Do not send duration or duration_seconds; Music 1.0 rejects them. Do not send a non-empty negative_prompt; it returns HTTP 400 with public_reason negative_prompt_unsupported, so put exclusions in the positive prompt. The prompt maximum is 5000 characters, and image_url, if used, must be a public HTTPS URL.

For length, say it in the prompt, for example a 2-minute track, or use section markers like [0:00-0:30] Intro: followed by the direction for that section.

Where does the finished file come from?

Submit to POST /v1/music-router/generate, poll the job, then read the audio artifact from result.artifacts where type is audio. Sume hosts the file on media.sume.com; use that URL, not a provider link.

How long should the prompt be?

Long enough to hold all seven axes and short enough to read in one pass. The limit is 5000 characters, but a brief of three to five sentences usually carries all seven. Put the arc last, because it is the part you will revise most, and keep a copy of every version next to the file it produced.

If a result misses, change one axis at a time. Changing tempo, key and instruments together makes the next take impossible to compare with the last.