Google Lyria 3.5: Free AI Music Generation in Gemini, Explained
Google has folded its most advanced music-generation model straight into the Gemini app, for free, for every user on the planet. As of September 4, 2026, Lyria 3.5 lets anyone type a prompt like "an upbeat acoustic song about starting a new job" and get back a full track, with singing or as a pure instrumental, no music production skills required. We explain what it actually does, how to try it, and where the guardrails are.
⚡ Quick facts
- Rollout date: September 4, 2026, in the Gemini app and Gemini API; also in Flow Music, AI Studio and Google Vids
- Who can use it: Google says all Gemini users globally, on the web and in the app; you must be 18 or older, and daily limits can vary by plan
- Song length: Up to about 3 minutes, with vocals or as an instrumental
- Labels: every track carries an inaudible SynthID watermark, and Gemini can check an uploaded file for one
What Lyria 3.5 actually is
Lyria is Google DeepMind's family of AI music models, and 3.5 is described by Google as its "best-sounding" version yet, with more expressive, emotionally nuanced vocals, richer instrumental arrangements, better pronunciation, and tighter adherence to what you actually asked for in a prompt. It is also available in Google Flow Music, the company's dedicated tool for artists and creators. The September 4 update is what makes it mainstream: the same model, minus the separate app, sitting inside the Gemini chat interface millions of people already use.
How to actually make a song with it
Inside the Gemini app or at gemini.google.com, you can describe a genre in your own words or pick one from a list, choose whether you want vocals or an instrumental-only track, and set how long the song should be, short or full-length. Google has also added ready-made templates to help people who don't know where to start, for things like background music for a video or a quick birthday song for a friend. You can prompt with text, and developers get the same model through the Gemini API, as Google's announcement notes. Google's music generation overview says tracks run up to about three minutes and that the feature is for users 18 and over in countries where the Gemini app is available.
It's not just in Gemini
The same model now sits behind four different Google products at once: the Gemini app and website for everyday users, Google Flow Music for artists and creative professionals, Google AI Studio and the Gemini API for developers building their own music tools, and Google Vids for anyone scoring a video inside Google Workspace. That's a wider spread than most Google AI features get on day one, and it signals Google sees AI music less as a novelty and more as an ingredient app-builders should have access to alongside image and video generation.
Our take: how to write a prompt that gets a usable track
Short prompts like “a happy song” leave every choice to the model. We get more usable results by naming five things: the use, the genre, the tempo or energy, the instruments, and the vocal. A worked example for a 30-second product video:
- Weak: “upbeat music for my video”.
- Better: “Instrumental background track for a 30-second cooking video, light acoustic pop, around 100 BPM, ukulele and hand claps, no vocals, ends on a clean final chord.”
Before you publish anything you made:
- Listen to the full track. AI music can drift in the second half or end abruptly; generate again rather than editing a bad ending.
- Check any lyrics. Sung words can be slightly garbled, so read them back if the track has vocals.
- Read the platform rules. YouTube, Instagram and stock sites have their own rules for AI-generated audio, and some ask you to label it.
- Keep it original. Describe a style, not a named artist or song; that keeps you clear of the guardrails and of copyright trouble.
The guardrails: no artist clones, no copyrighted lyrics
Google has built in specific limits rather than leaving Lyria 3.5 wide open. It reportedly refuses prompts asking it to imitate a named, real artist's voice, and it blocks requests to reproduce copyrighted song lyrics. Every track it generates also carries SynthID, Google's inaudible watermarking technology that lets the company (and compatible detection tools) identify a clip as AI-generated even after it's been shared, re-uploaded, or edited. It's the same watermarking approach Google already applies to Gemini's AI images and video, and you can upload an audio file and ask Gemini whether it carries the mark, Android Authority reports.
Does it work in India?
Yes. Google's own announcement says Lyria 3.5 is available "to all users globally on the web and in the mobile app," with no country carve-outs mentioned, so it works through the same Gemini app already installed on phones across India, alongside features the site has covered before like NotebookLM's Hindi and Tamil audio overviews and the recently added "Live" voice features in Gmail, Docs, and Keep. As with those, exact regional pricing or plan-tier gating can sometimes differ in ways Google doesn't spell out in its initial announcement, so if a specific option looks locked, that's worth double-checking inside your own Gemini app rather than assuming it's unavailable outright.
Sources
- Google: better tracks with Lyria 3.5 in the Gemini app (September 4, 2026)
- Gemini: music generation overview
- Android Authority: latest Lyria model comes to the Gemini app
- 9to5Google: Lyria 3.5 rolls out in Gemini
Frequently asked questions
How can I tell if a song was made with Lyria?
Check for SynthID. Google adds an inaudible SynthID watermark to every Lyria track, and you can upload an audio file to Gemini and ask whether it was made with Google AI. A missing mark does not prove a human made it, because other music tools do not use SynthID.
What is Google Lyria 3.5?
Google DeepMind's latest AI music-generation model, now powering a song-making tool built into the Gemini app and Gemini API since September 4, 2026, that turns a text prompt or image into a full track with vocals or as an instrumental.
Is Lyria 3.5 free, and does it work in India?
Google says Lyria 3.5 is available to all Gemini users globally, on the web and in the mobile app, including India. You must be 18 or older, and how many tracks you can make a day can depend on your Google AI plan.
How long can a Lyria 3.5 song be, and what can I control?
Songs can run up to about 3 minutes with real verse, chorus, and bridge structure. You can pick or describe a genre, choose vocal or instrumental, set the track length, and use built-in templates to get started.
Can I clone an artist's voice or use copyrighted lyrics with Lyria 3.5?
Reportedly not. Google's guardrails are said to block prompts that imitate a specific real artist's voice or reproduce copyrighted lyrics, and every generated track carries an inaudible SynthID watermark identifying it as AI-made.