Gemini can make music now. Describe what you’re after — a genre, a mood, an odd inside joke between you and your buddies — and it spits out 30 seconds of song. Lyrics included for the taking, instrumental for everyone else.
The feature operates on Lyria 3, the latest generative music model by Google DeepMind. Now it’s live as a beta today in the Gemini app, and it will be released across eight languages: English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese. It’s available to use if you’re over 18, and if you have a plan (AI Plus, Pro, or Ultra), there are higher usage caps than for free users.
How It Works
Pretty straightforward. Type a prompt describing what you’re after — could be a genre, a feeling, a specific memory, whatever — and Gemini builds a track around it. You can also upload a photo or video and tell it to pull inspiration from that instead. The cover art gets auto-generated too, by something called Nano Banana, which is a name that raises more questions than it answers, but there you go.
Lyrics are created automatically as part of the process. You don’t need to write them yourself, though that does mean you’re trusting an AI to put words to your half-formed musical idea. Make of that what you will.
What Google Says It’s For
Google is being pretty upfront that this isn’t meant to produce actual songs you’d stick on a playlist. Their words: the goal is to give people “a fun, unique way to express themselves” and share the results with friends. So think novelty clips for group chats, personalised birthday jingles, that kind of thing. Not your next album.
If you name a specific artist in your prompt, Gemini won’t try to copy them directly. Google says it treats artist names as broad creative direction — similar style, similar mood — rather than attempting an imitation. Whether that distinction holds up in practice is another question entirely, but that’s the stated intent.
The Copyright and Watermark Stuff
Every track Gemini generates is tagged with SynthID, Google’s invisible watermark for flagging AI-generated content. You won’t hear it or see it, but it’s baked into the file so the origin can be identified later.
Google also says there are filters checking outputs against existing music to avoid accidentally reproducing something that already exists. And if something does slip through — or if you spot a track that sounds suspiciously close to someone else’s work — there’s a reporting system for flagging potential rights violations.
That last bit matters more than it might seem. Generative music is heading straight into the same legal grey zone that AI image generation has been stuck in for the past couple of years. Google clearly knows this, which is probably why they’re emphasising original expression and building in safeguards from the start rather than bolting them on after the first lawsuit.
