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

Google DeepMind launched Lyria 3.5 in Google Flow Music on Jul 29, 2026. The model is positioned as the newest music generation system from the lab, with stated gains in musicality, lyrics, vocals, and creative control. It is available the same day inside Flow Music, which Google frames as a place to create songs with direct creative control rather than one-shot generation alone.

On the product side, Lyria 3.5 is described as producing richer, more complex melodic structures that sound more natural. Lyrics generation is claimed to improve quality through stronger prompt adherence and structural awareness, so song form and wording track the user’s request more closely. Vocals are said to carry more expression and emotion, with more realistic, emotionally nuanced delivery and better pronunciation. Creative control is expanded around tempo and duration, so users can shape length and pacing of outputs more easily inside Flow Music.

For engineers and builders, the useful signal is not a new API surface in this announcement but a clearer product stack: a named generation model (Lyria 3.5) paired with an app surface (Flow Music) that exposes tempo and duration as first-class controls. Prompt adherence and structural awareness for lyrics matter if you are evaluating music models for workflow tools, demos, or creative pipelines where outputs must stay on brief. Vocal realism, pronunciation, and emotional nuance matter for any use case where sung performance is the product, not just instrumental texture.

In market terms, Google is shipping the model through Flow Music rather than as a standalone research drop, and is selling control—tempo, duration, and tighter lyric structure—alongside raw generation quality. The pitch is incremental capability inside an existing creation product: richer melodies, better lyrics, more expressive vocals, and knobs for pacing and length. That positions Lyria 3.5 as a product upgrade path for users already in Flow Music, not only as a model card update.

Practical next step: try Lyria 3.5 in Flow Music and stress-test the four claimed axes. Prompt for multi-section songs and check whether lyrics hold structure and stay on brief. Push emotional range and hard-to-pronounce lines to judge vocal quality. Vary tempo and duration on the same prompt to see how much control you actually get. Watch whether those controls stay limited to tempo and duration or expand as Flow Music evolves.