Google just handed the mic to Lyria 3.5. On September 4, the company announced that its most advanced music model is now built into the Gemini app and available through the Gemini API — which means the thing Google calls its "best-sounding" music AI is suddenly sitting in front of hundreds of millions of people.
What changed
Lyria 3.5 had been living in the creative corners of Google's ecosystem. It debuted earlier this year in Google Flow Music, the company's tool for artists and AI-first musicians, where most Gemini users would never stumble across it. That's over. Anyone on Gemini — web or mobile — can now generate complete tracks, and developers can wire the same model into their own products through the API.
Google says the DeepMind-built model produces more expressive vocals and richer arrangements than earlier Lyria versions, with a real jump in fidelity. In plain terms: tracks that sound less like a demo loop and more like finished music.
How it works in the app
You don't need to know what a DAW is. Pick a genre from a list, or just describe one. Decide whether you want vocals or an instrumental. That's the workflow.
There are also new templates that set the model up for common jobs — background music for a video, a jingle for a small business, a custom birthday song. And you can finally choose between a short clip and a longer, fuller track, which had been one of the more annoying limits of earlier in-app music experiments.
Google's own examples say a lot about who this is for: video creators who need backing tracks, small businesses that want a jingle, and anyone who's ever wanted their own ringtone.
Why it matters
Music has been the slowest generative-AI medium to reach regular people — partly a quality problem, partly the legal minefield around training data and sound-alike vocals. By shipping Lyria 3.5 inside Gemini, the same app where people already draft emails and generate images, Google is betting that music generation is ready to be a default feature instead of a specialist tool.
It changes the competitive math too. AI-music startups have built real audiences on one-click songwriting, and other labs keep circling audio. Google's edge is distribution. Gemini's install base dwarfs any standalone music app, and the API release means Lyria 3.5 can end up inside third-party video editors, games, and marketing tools without users ever learning its name.
For working musicians, it's more complicated. Cheap, fast, royalty-free background music is exactly the segment that pays session players and stock-music composers right now. A model that turns out a serviceable jingle in seconds will squeeze that market long before it troubles anyone at the top of the charts.
Where this goes
The rollout is global and immediate, and the model is also live in Google AI Studio and in Google Vids, the company's workplace video tool. So this isn't a consumer experiment — Google wants music generation running across consumer, developer, and enterprise products at the same time.
The unresolved question is what happens when this much music-making capacity meets platforms already flooded with synthetic audio. Streaming services have started labeling and throttling AI tracks, and rights holders are suing over training data across the industry. Google shipped the instrument. The rules for playing it are still being written.
Image: cottonbro studio, via Pexels
