Google brings Lyria 3 to Gemini for instant 30-second music tracks

Instead of sending a photo or a meme to your group chat, Google now wants you to send a custom 30-second song with auto-generated cover art.

Google just rolled out Lyria 3, DeepMind’s latest music generation model, in beta directly inside the Gemini app. The feature is available to users 18 and older on desktop today, with mobile access rolling out over the next few days.

Photos, prompts, and automatic album art

Lyria 3 generates 30-second tracks based on text prompts, photos, or uploaded videos. You can ask for a specific genre, tempo, and mood, or simply drop in a picture of your dog on a hike and ask Gemini to set the scene to music.

One big leap over earlier models is that you no longer need to write lyrics. Gemini writes and sings them based on your prompt. To make sharing frictionless, the app also uses Nano Banana to generate a matching custom album cover for every clip.

Outside of the Gemini chatbot, Google is also rolling out Lyria 3 to YouTube’s Dream Track tool across more countries, giving Shorts creators a native way to generate background audio.

At launch, it supports eight languages, including English, Spanish, French, German, Hindi, Japanese, Korean, and Portuguese. Subscribers to Google AI Plus, Pro, and Ultra tiers get higher usage limits.

Guardrails up, watermarks embedded

Generative audio has faced heavy pushback from the music industry, and Google is taking a visibly cautious approach with this release.

DeepMind first introduced Lyria back in 2023, where early demos included cloning the voices of famous artists. With Lyria 3 in Gemini, that door is firmly shut. Google says the model is designed for original expression rather than mimicry; naming a specific artist will only influence the general genre or mood, backed by automated filters that check against existing copyrighted tracks.

Every track generated in the app carries Google’s invisible SynthID watermark. Gemini can now also inspect uploaded audio files to verify whether they were generated by Google AI.

Why it matters

Google is arriving late to the consumer AI music party, trailing behind standalone hits like Suno and Udio as well as Microsoft’s Copilot integration. But it is taking a fundamentally different strategic angle.

Instead of pitching this as a serious composition studio, Google is treating audio as another casual messaging format. For everyday users, 30 seconds is not enough to produce a radio hit, but it is the perfect length for a quick laugh, a birthday greeting, or a YouTube Short soundtrack. It turns music creation into a low-stakes social tool.

We think keeping tracks strictly capped at 30 seconds is a clever engineering choice. It sidesteps the hardest challenge in generative audio, which is keeping long-form musical structures coherent, while delivering instant gratification.

The real moat here is YouTube. If Google can turn Lyria into the default soundtrack engine for millions of Shorts creators, it won’t need to compete head-on with pro-audio platforms to capture massive consumer volume.