Google DeepMind released Lyria 3.5, its newest music generation model, on July 29, 2026, and is rolling it out inside Google Flow Music. The update targets four axes at once — musicality, lyrics, vocals, and creative control — and lands as the generative-audio field shifts from novelty demos to tools creators actually keep open next to a DAW. Google is positioning Lyria as the production endpoint rather than a research artifact.
The most consequential change is on lyrics. Previous Lyria releases could produce singable text but struggled with prompt adherence — a user asking for a bridge about a specific subject often got a chorus about something adjacent. Lyria 3.5 adds what DeepMind describes as structural awareness, meaning the model tracks verse, chorus, and bridge boundaries against the prompt.
“Our newest music generation model, Lyria 3.5, delivers significant advancements across musicality, lyrics, and vocal quality, empowering you to craft richer tracks.”— Google DeepMind, Lyria team blog post
Vocals get the second-largest push. DeepMind cites more realistic and emotionally nuanced delivery plus improved pronunciation — the failure mode where a generated vocalist mangles a proper noun or slurs a consonant. That has been the tell for AI-generated tracks in blind tests, and closing it narrows the daylight between Lyria output and demo-quality human vocals.
Key facts
- 01Google DeepMind released Lyria 3.5 on July 29, 2026, rolling out through Google Flow Music.
- 02The update targets four axes: musicality, lyrics, vocals, and creative control including tempo and duration.
- 03Lyric generation now handles prompt adherence and structural awareness rather than free-form text over a melody.
- 04Vocal output adds emotional nuance and improved pronunciation, a persistent weak spot in generative music.
- 05Creators can now more directly steer tempo and clip length, closing the gap with traditional DAW workflows.
Musicality — the catch-all for melody, harmony, and arrangement — gets richer melodic structures that DeepMind says sound more natural. This is the axis where generative music models have historically produced technically correct but emotionally flat output: chord progressions that resolve but don't build. Whether Lyria 3.5 clears that bar depends on ears, not benchmarks, and the audio community will judge quickly.
Creative control is the workflow story. Users can now more directly steer tempo and duration of outputs, two controls that sound trivial but that have blocked professional adoption. A composer scoring a 32-second ad break does not want a 45-second track they have to trim; a producer building a house track wants 124 BPM, not something close to it.
Flow Music is the surface, not a separate model. Google is treating Flow as the creator-facing product where Lyria capabilities ship first, similar to how ChatGPT is the front door for OpenAI's model releases. That decoupling lets DeepMind iterate on the model without breaking the product, and lets Flow accumulate workflow features — arrangement, editing, revision history — independently.
The competitive backdrop matters. Suno and Udio have built consumer momentum in AI music, and both face active copyright litigation from the major labels over training data. Google has not disclosed Lyria 3.5's training corpus in this release, and the company has previously said it uses licensed and proprietary content for its music models. That posture is a hedge against the legal exposure smaller competitors are now absorbing in court.
DeepMind did not publish benchmark scores or listener-preference data alongside the release, which is unusual for a model launch and leaves the improvements to be judged by creators in the tool. There is also no pricing disclosure for Flow Music tiers in the announcement, and no word on API access for developers who might want Lyria 3.5 outside the Flow product.
The bet Google is making with Lyria is that generative music becomes a professional creator tool rather than a party trick, and that the winner is whoever can ship the vocals, lyric coherence, and DAW-compatible controls that let a working musician use the output without re-recording it. Lyria 3.5 pushes all three at once, which is either a coordinated product cycle or a sign that the previous version was further from professional-usable than the marketing suggested — probably both. The interesting number to watch is not a benchmark but how many Flow Music tracks show up in released commercial work over the next two quarters.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
