Lyria 3.5 prompts: [Verse] tags or [0:00-0:30] time ranges?

Google's Lyria 3.5 docs name section tags in the prompt: [Verse], [Chorus] and [Bridge]. Sume's music docs show a different device for the same model family: time-range markers such as [0:00-0:30] Intro: .... Sume's docs do not describe the Google tags, so treat them as something to test rather than assume.

Both are plain text inside one prompt, so a side-by-side test costs two small jobs.

What each page says

Only statements from the two pages as read on 2026-10-03.

Two versions of one brief

The brief below is for a 45-second product video bed. Version A uses the tags from Google's docs. Version B uses the time ranges from Sume's. Both keep the same instruments, tempo and closing clause so the structure device is the only variable.

A (section tags)
Warm lo-fi hip hop, 84 BPM, C minor. Dusty Rhodes, brushed drums, muted trumpet.
[Verse] Sparse Rhodes and brushes, no trumpet.
[Chorus] Trumpet answers the Rhodes, bass thickens.
[Bridge] Drums drop out, Rhodes alone.
Instrumental, no vocals.

B (time ranges)
Warm lo-fi hip hop, 84 BPM, C minor. Dusty Rhodes, brushed drums, muted trumpet. A 45-second track.
[0:00-0:15] Intro: sparse Rhodes and brushes, no trumpet.
[0:15-0:35] Main: trumpet answers the Rhodes, bass thickens.
[0:35-0:45] Outro: drums drop out, Rhodes alone.
Instrumental, no vocals.

How to compare them fairly

Submit each version through POST /v1/music-router/generate and listen with the brief in front of you.

• Run each version at least twice. Google's docs say the model is single-turn, so there is no edit step to correct a miss; a miss means a new job.
• Check the section boundaries by ear against the markers. Sume's music docs say a brief is a creative direction, not a guaranteed output setting, so verify the audio.
• Read result.lyrics on the finished job if present. The docs call it model-reported metadata describing structure, not an audio measurement.
• Keep the closing clause identical. Sume's docs put exclusions in the positive prompt, for example "Instrumental, no vocals.", because a non-empty negative_prompt returns a 400.

Which to default to

On Sume, default to the time-range form, since it is the one Sume's docs show and it doubles as a length statement. If you also render the same brief directly against Google's API, keep the tag form for that path and keep both versions in your prompt library under one brief id. The goal is a result you can reproduce from a stored prompt, not a rule about brackets.