Flow Music adds BPM control with its Lyria 3.5 upgrade

Flow Music now runs on Lyria 3.5, Google DeepMind's newest music generation model, and the upgrade targets the three things that most separate AI-generated music from something a human made: how the vocals sound, how the arrangement holds together, and how closely the output follows what you actually asked for.

More expressive vocals

Lyria 3.5 targets more expressive, dynamic singing, the difference between a vocal line that sounds technically correct and one that carries actual performance, dynamics, phrasing, emotional lift on a chorus. That's historically been one of the harder problems in AI music generation, since a flat or robotic vocal is often the single biggest tell that a track was AI-made.

Arrangements that flow, not just loop

The model also brings richer arrangements, with tracks described as flowing naturally rather than sounding like a loop stitched into a longer runtime. That points at better structural coherence across a full song, verses, builds, and choruses that connect logically instead of a beat repeating with minor variation.

New BPM control

On the direction-following side, Lyria 3.5 adds a dedicated BPM control, letting creators lock in a specific tempo rather than describing it in words and hoping the model lands close. That's a meaningful practical upgrade for anyone syncing music to video, matching a track's tempo to an edit's pacing is far easier when BPM is a direct parameter instead of a fuzzy prompt instruction.

Where this fits in Flow Music's roadmap

The model upgrade follows Spaces' recent expansion into a full vibe-coding environment for building custom instruments and music apps, generating, editing, and stemming songs all inside one tool. A stronger underlying model raises the ceiling on everything built on top of it, from quick song generation to the more involved Spaces workflows.

Flow Music

Google's AI music generation tool for creating and editing full songs.

View Flow Music →

Sources: Flow Music on X.