Google launches Lyria 3.5 in Flow Music with better melodies, lyrics, and vocals

Google DeepMind has released Lyria 3.5, the latest version of its music-generation model, rolling it out today inside Google Flow Music. The update focuses on four areas of the music-creation pipeline: melodic structure, lyric generation, vocal realism, and control over tempo and duration.

What's new

Google's announcement lays out specific gains over prior versions across each dimension of song creation:

• Musicality — the model can "create richer, more complex melodic structures that sound more natural."
• Lyrics — Lyria 3.5 can "generate higher quality lyrics with improved prompt adherence and structural awareness," meaning generated lyrics should track more closely with what a user actually asked for and hold together structurally (verse/chorus form) rather than drifting.
• Vocals — the model adds "more expression and emotion to your songs with more realistic and emotionally nuanced vocals, plus improved pronunciation."
• Creative control — users can "more easily control the tempo and duration of your outputs," giving finer-grained direction over a track's length and pace rather than accepting whatever the model defaults to.

Google says the model is available "today in Flow Music." The company's post does not specify pricing, regional availability, or eligibility requirements for access.

Context

Lyria is Google DeepMind's text-to-music model family, and Flow Music is the music-generation surface inside Google's Flow creative suite, sitting alongside Flow's video and image tools. Positioning Lyria updates inside Flow keeps music generation next to Google's other generative media tools rather than as a standalone product, letting a creator building a video project reach for a matching soundtrack in the same workspace.

The release lands in a crowded month for AI music: independent labs like Suno and ElevenLabs have been pushing on vocal realism and songwriting control from their own angles, and improvements in lyric adherence and expressive vocals are the two areas where competitors most often criticize generative music tools — models either sound robotic or ignore the prompt's structure. Google's framing of this release speaks directly to both weaknesses.

Why it matters

The specific gains Google is claiming — better prompt adherence for lyrics, more natural vocal expression, and finer tempo/duration control — are the details that separate a novelty demo from a tool a musician or video creator would actually use for finished work. Structural drift in AI-generated lyrics and flat, mechanical-sounding vocals have been the most common complaints leveled at earlier generations of text-to-music models, including Lyria's own prior versions and rivals' offerings.

By shipping the update directly into Flow Music rather than as a separate release, Google is also signaling how it wants generative music to be used: as one layer of a broader video/creative pipeline, not a standalone jukebox. That matters competitively, since it ties music generation to Flow's existing video and image tools, giving Google a bundled offering to set against point-solution music generators.

No pricing or availability details beyond "today" were disclosed, so it remains to be seen whether Lyria 3.5 is a broad rollout to all Flow users or a staged release — a detail worth watching as adoption data emerges.