Google ships Lyria 3.5 in Flow Music with vocal, tempo upgrades

A new music model shipping straight into a consumer app, with no benchmarks and no comparisons, is the kind of release that used to feel underwhelming and now feels normal. Google rolled out Lyria 3.5 inside Google Flow Music on July 29, pitching four improvements: richer, more complex melodic structures, higher-quality lyrics with better prompt adherence and structural awareness, more expressive and emotionally nuanced vocals with improved pronunciation, and easier control over tempo and duration of outputs. The post is credited to Google DeepMind.

The interesting part is what is missing. The company put no numbers next to any of those claims, no head-to-head against prior versions, no listener-panel data, no comparison against rival music generators. Take the specifics as reported, not settled. The model is only exposed inside Flow Music, so there is nothing an outside researcher can independently benchmark right now.

Why this matters if you are not building song generators: control features are becoming the frontier, not raw fidelity. Tempo, duration, and pronunciation are the boring plumbing that decides whether a generated track can slot into an actual creative workflow versus stay stuck as a novelty. Google is competing on usability, not just spectacle, and that is a real signal about where the music-generation category is heading.

The honest caveat is that a company post is a marketing document. What the reporting does not give you is any word on provenance, watermarking, or artist-opt-out for the new vocal capabilities, and that is exactly where labels, publishers, and rights-holders will push next. Better vocals are also better impersonation.

The upside, if the model lands as pitched, is more control for the hobbyist and indie creators inside Flow Music, where a tighter loop from prompt to usable track matters more than one more decibel of fidelity.