We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control

Source: DeepMind Date: 2026-07-29 URL: https://deepmind.google/blog/were-launching-lyria-35-in-google-flow-music-with-advances-across-musicality-lyrics-vocals-and-creative-control/

Summary

Google DeepMind shipped Lyria 3.5, an incremental update to its music-generation model, rolling out in Google Flow Music. Claimed gains are user-facing: richer/more natural melodic structure, better lyric prompt-adherence and structural awareness, more emotionally nuanced and better-pronounced vocals, and finer tempo/duration control. No architecture, parameter count, or training details were disclosed.

Implications

Feeds the generative-media thread as the latest point release in Google’s music-gen lineage (Lyria 3.5 following 3.x) — evidence that music generation is iterating on the same “point-release polish” cadence as image/video models, with creative-control knobs (tempo, duration, lyric adherence) now a competitive axis alongside raw output quality. No technical disclosure means it can’t be cross-compared against other music models (Suno, etc.) beyond marketing claims.