Google’s Gemini's Lyria 3 AI music tool lets you create songs from text or photos

Google has rolled out music generation in the Gemini app, powered by its latest DeepMind model Lyria 3. Users can now generate 30-second tracks using a text description, an image, or a video clip—no musical knowledge required. The feature is available in beta to all Gemini users aged 18 and older across eight languages: English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese. Free users get a taste of the feature, while Google AI Plus, Pro, and Ultra subscribers get higher usage limits. Paid users will be able to push it further, though Google hasn't specified by exactly how much.

Lyria 3 writes its own lyrics—you just set the vibe

Unlike earlier versions of the model, Lyria 3 generates its own lyrics based on your prompt. You no longer have to supply them yourself. Users can describe anything—a genre, a mood, a memory, an inside joke—and the model handles the rest, including elements like tempo, vocal style, and instrumentation. Each track also gets custom cover art generated by Google's Nano Banana image model, making it easy to share directly from the app.

Prompts can get surprisingly specific. Google's own example asks for "a fun afrobeat track with a true African vibe" as a tribute to a mother's home-cooked plantains. You can also go the other route: just drop in a photo of your dog on a hike and let Gemini compose something that fits the moment.

Google uses SynthID to watermark every AI-generated track

To address authenticity concerns, all Lyria 3 outputs are embedded with SynthID, Google's imperceptible AI watermark. You can upload any audio clip to Gemini and ask whether it was AI-generated—the app will check for the watermark and flag it accordingly.

On the copyright front, Google says Lyria 3 is built for original expression rather than artist mimicry. Naming a specific artist in a prompt won't clone their sound—instead, Gemini treats it as "broad creative inspiration." Filters are also in place to screen outputs against existing content, though Google acknowledges the system isn't perfect and encourages users to report potential violations.

The feature is live on desktop now, with the mobile app rollout following over the next few days.