Lyria 3 Pro: Create longer tracks in more
Google DeepMind
Google DeepMind's Lyria 3 Pro can now generate music tracks up to 3 minutes long, with control over intros, verses and choruses. It's rolling out across Vertex AI, Gemini, Vids and more, putting AI music tools in front of way more people at once.
Based on reporting by Google DeepMind — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google DeepMind just widened the aperture on its music generation model. Lyria 3, which launched only last month, gets a Pro upgrade that stretches maximum track length out to three minutes and adds something more interesting than duration: actual structural awareness. You can now prompt for a bridge, a verse, a chorus, and the model understands those aren't just words but functional parts of a song that need to connect to each other.
That's the real story here, not the runtime. Anyone can pad out a generation. Getting a model to grasp that a chorus should feel different from a verse, and that a transition between the two needs its own logic, is a harder problem, and it's the piece that determines whether AI-generated music sounds like a composition or a loop that got lucky.
Google isn't keeping this behind one gate. Lyria 3 Pro is now in public preview on Vertex AI for companies that want on-demand audio at scale, whether that's game studios needing bespoke soundtracks fast or platforms bolting music generation into their own tools. Developers get it through AI Studio and the Gemini API alongside Lyria RealTime. Google Vids, the AI video app, is folding it in for Workspace customers and AI Pro and Ultra subscribers starting this week, so marketing videos and creative projects get custom scoring without a licensing headache. Paid Gemini app subscribers get the longer generations too, useful for anyone scoring a vlog, podcast, or tutorial. And ProducerAI, the collaborative songwriting tool Google introduced recently, now runs on Lyria 3 Pro for both free and paid users globally, offering something closer to an agentic co-writer than a one-shot generator.
On the responsibility front, Google is leaning on its Music AI Sandbox work with actual musicians, including a collaboration with Grammy-winning producer Yung Spielburg on a DeepMind short film score, plus DJ and producer François K working the model into a forthcoming track. The company says Lyria 3 won't mimic a named artist directly, treating any such prompt as loose inspiration rather than impersonation, and every output carries a SynthID watermark so AI-generated tracks stay traceable. Training data comes from content YouTube and Google already have rights to use, which is the company's way of preempting the copyright fight before it starts.
Whether that's enough guardrail for an industry still nervous about AI encroaching on songwriting is a separate question. But the rollout itself, five different products getting the same model on the same week, tells you Google isn't treating music generation as a side experiment anymore. It's infrastructure now.
My take — AI-written commentary, not fact-checked reporting
The three-minute cap and the SynthID watermarking tell you exactly how nervous Google still is about copyright blowback, and rightly so. I'd rather see this land as a genuine songwriting collaborator, like the ProducerAI angle suggests, than as a royalty-free stock-music vending machine that quietly guts session musicians and library composers who used to make a living scoring exactly this kind of marketing filler.
Read more about this at: Google DeepMind