Lyria 3.5 tempo and duration control: Flow Music versus the Sume API

Google says Lyria 3.5 gives easier tempo and duration control in Flow Music. On Sume the same model has no duration field, so you steer length in the prompt.

4 min readSume
All posts

Google's Lyria 3.5 announcement lists creative control as one of four advances: you can more easily control the tempo and duration of your outputs in Flow Music. Sume's Music 1.0 runs on Lyria 3.5 too, but it has no duration or tempo parameter. Tempo and length live in the prompt, and a request that sends duration or duration_seconds is rejected.

Where the control lives

The Flow Music announcement describes an app feature, and the blog post does not say how the control is exposed or whether it is in an API. So do not assume a field you saw in an app exists on a REST route. In Sume the request has prompt (1 to 5000 characters), an optional image_url, metadata, and the usual mode and webhook_url fields.

What to write in the prompt

Write the tempo as a number, such as "72 BPM", and the length as a plain phrase, such as "a 2-minute track", or use section markers like [0:00-0:30] Intro: .... The Sume docs describe these as creative directions, not guaranteed output values, so check the length of the file you get back.

curl -X POST https://api.sume.com/v1/music-1.0/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: score-72bpm-001" \
  -d '{"prompt": "A 30-second neo-soul nocturne, 72 BPM, D minor. Rhodes and soft sub bass. Instrumental, no vocals."}'

How the two surfaces compare, read 2026-10-06:

Tempo and duration, Flow Music versus Sume, read 2026-10-06
QuestionFlow Music (Google blog)Sume Music 1.0 (docs)
ModelLyria 3.5Lyria 3.5
Tempo controlEasier control, per GoogleBPM written in the prompt
Duration controlEasier control, per GoogleNo field; named in the prompt
Negative promptNot statedNon-empty value returns 400
PriceNot stated$0.125 per generation

Checking what you got

Because the prompt only steers the result, measure it. Read the duration of the returned file, and tap out the tempo against a metronome or ask your editor to show the BPM. If the length is wrong, do not rerun at random. Change one phrase of the prompt, such as the section markers, and compare.

Provider lyrics on the result describe tempo and structure, but the docs call them model-reported metadata, not a measurement of the audio. Treat the file as the source of truth, and keep the take that fits the cut.

If the length must be exact, generate a bit long and cut it. A timeline render takes the music as a soundtrack with a fade_out_seconds of up to 10, so the output ends where the picture ends. Confirm the live rate in GET /v1/catalog.

Sources

Related posts

More in Models

All Models posts

Written by Sume