Lyria 3 is Google DeepMind's music generation model: describe a genre or upload a photo, and it sings back an original track, lyrics included. Vocals span eight languages, and every file carries a SynthID watermark. The feature runs in the Gemini app as a beta for adults, while the Pro variant stretches songs to three minutes through an API billed per generation.
Inside Gemini, you type a genre, a tempo and a mood, and the model returns three 30-second candidates, ready to save as MP3 or as a subtitled video. Skip the lyrics and it writes them for you, rhymes included, then performs them with vocals that hold up on repeat listens.
Pictures work as well as words. A sunset drifts toward ambient, a party shot toward something punchier (Google's own demo prompt, an R&B slow jam about a sock reunited with its match, sets the expected level of gravitas). Under the hood sits a latent diffusion model applied to compressed audio representations.
Lyria 3 comes in several flavors depending on length and channel. The Pro variant composes songs of roughly three minutes in 48kHz stereo, with intros, verses, choruses and bridges you can direct straight from the prompt, through the Gemini API and Google AI Studio.
The lineup moves fast, too: Lyria 3.5, unveiled since then, now tops the family inside Google Flow Music.
| Version | Output | Where to find it |
|---|---|---|
| Lyria 3 | Sung 30-second clips | Gemini app, Dream Track on YouTube Shorts |
| Lyria 3 Pro | Structured songs around 3 minutes, 48kHz stereo | Gemini API, Google AI Studio |
| Lyria RealTime | Live, on-the-fly music generation | API, for developers |
| Lyria 3.5 | Sharper musicality, lyrics and vocal quality | Google Flow Music |
The quickest way in is gemini.google.com/music, open to adult accounts with quotas tied to your plan. Google AI Studio lets you experiment for free before wiring up an API key, priced at $0.04 per Lyria 3 clip and $0.08 per Pro song. Those figures shift with Google's release cycle, so treat the official pricing page as the final word.
Across the street, Suno 5.5 courts hobbyist songwriters with a dedicated subscription, a segment our roundup of the best AI music generators digs into. Google aims elsewhere: personal, playful soundtracks rather than professional record production.
Partly. Music generation ships inside the Gemini app for accounts aged 18 and over, with limits tied to your plan, and Google AI Studio allows free experiments. Heavier volumes go through the API, which charges per generated track rather than through a flat subscription.
Two philosophies. Suno chases the complete song through its own subscription service, while Lyria 3 leans on Google's own products: Gemini, YouTube Shorts and the API. For a quick jingle without learning yet another tool, DeepMind's model already lives in the app you chat with daily.
It depends on the access point. Google frames the Gemini feature as personal expression, while some partner platforms list their Lyria 3 output as royalty-free for commercial projects. Read the terms of the specific service you generate from before publishing a monetized video.
Every generated file embeds SynthID, an inaudible digital watermark developed by Google DeepMind. It flags the audio as AI-made without changing how it sounds, and Google supplies a verification tool so you can check the origin of a suspicious file.
Verdict: Suno courts the bedroom songwriter; Lyria 3 slips into the app you already talk to. Podcasters, video makers and anyone plotting a musical inside joke get a pocket studio right inside Gemini.
