On March 25, Google DeepMind shipped Lyria 3 Pro, its most advanced AI music generation model to date. A single generation now runs up to three minutes of finished music — a hard ceiling break from Lyria 3, which debuted only a month earlier with a 30-second limit. The same update pushed the model into the Gemini app, Google Vids, Vertex AI, AI Studio, and ProducerAI, the agentic music production tool Google acquired in February. In one move, AI music went from experimental snippets to deliverable-length output.
For developers, the real signal is not that a model can compose. It is that music generation now has a first-class seat in Google’s API and enterprise channels. With a public preview on Vertex AI and model access through the Gemini API, audio generation is being sold as a billable, scalable primitive — the same tier of infrastructure as text and images.
From 30 Seconds to 3 Minutes: Structured Generation
Lyria 3 Pro does not get to three minutes by stitching clips together. Google says it has a deeper understanding of musical composition: you can prompt for specific elements like intros, verses, choruses, and bridges, and the model handles the transitions between sections and experiments across styles. TechCrunch described the jump from Lyria 3’s 30-second cap as a major one.
The workflow consequences are concrete. A 30-second output is a demo or a sample; three minutes covers a short video, a podcast intro, or in-game scene scoring — the exact use case Google named for the Vertex AI preview. The model was also shaped with working musicians through the Music AI Sandbox: Grammy winner Yung Spielburg scored the DeepMind short film “Dear Upstairs Neighbors” with Lyria, and DJ and producer François K used it for an upcoming release.
Rollout Across Products and APIs
The distribution surface is worth listing in full:
- Vertex AI: public preview for businesses generating audio on demand and at scale
- AI Studio and the Gemini API: developers can pair Lyria 3 Pro with Lyria RealTime to build real-time creative tools
- Google Vids: rolling out that week to Workspace customers and Google AI Pro and Ultra subscribers
- Gemini app: longer generations for paid subscribers, with vlogs, podcasts, and tutorials named as targets
- ProducerAI: the agentic production tool acquired in February, open globally to free and paid users
The Gemini API changelog listed lyria-3-pro-preview the same day, confirming the API path went live alongside the consumer surfaces. Products first, API right behind — the same cadence as the model acceleration we tracked in our opening-of-2026 outlook, except this time what is accelerating is audio.
Training Data and the Licensing Line
No music model escapes the data question. Google’s position: training used materials that YouTube and Google have a right to use under their terms of service, partner agreements, and applicable law. The models “do not mimic artists” — naming a creator in a prompt works only as broad stylistic inspiration — and outputs pass through filters that check them against existing content. Every Lyria 3 and Lyria 3 Pro track carries a SynthID watermark identifying it as AI-generated.
The industry context makes the caution deliberate. Spotify has shipped tools letting artists review songs released under their name, and Deezer built detection tooling other streaming services can use to flag AI-generated music. Google’s bundle — licensed-or-permitted data, watermarking, no artist mimicry — reads as a clear attempt to stay out of the litigation zone where Suno and Udio have spent the past two years.
What It Means for Developers and Creators
Three takeaways. First, once audio generation is an API, its cost structure becomes comparable: game soundtracks, video scoring, and podcast beds — the “good enough” tier of audio — will be priced against generation costs first. Second, SynthID shipping as a factory default means detection and disclosure of AI audio is becoming platform-level infrastructure, not a per-company policy choice. Third, opening ProducerAI to free users worldwide is both acquisition and moat: once a producer’s sample library and workflow live inside Google’s ecosystem, switching costs only compound.
Sources
- Google launches Lyria 3 Pro music generation model — TechCrunch
- Lyria 3 Pro: Create longer tracks in more Google products — Google Blog
- Gemini API Release Notes — Google AI for Developers
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.