Google launched Lyria 3 Pro on March 25, a significant upgrade to its AI music generation model released just a month ago. The most immediate change: tracks can now run up to three minutes long, compared to the 30-second limit in the original Lyria 3.

Beyond length, Lyria 3 Pro gains what Google calls an understanding of "music architecture." Users can now specify structural elements in their prompts — intros, verses, choruses, and bridges — giving composers more deliberate control over how a song unfolds. Google says transitions between sections are more natural and complex than the previous model.

The rollout spans several surfaces:

  • Gemini app — available to paid (Pro/Ultra) subscribers
  • Google Vids — for Workspace customers and Gemini paid users
  • ProducerAI — the AI music production tool Google acquired last month
  • Vertex AI (in public preview), Gemini API, and Google AI Studio for developers

On the training data question, Google says Lyria 3 Pro was built on partner-licensed content plus permissible data from YouTube and Google properties. It does not attempt to replicate individual artists, though if an artist is named in a prompt the model takes "broad inspiration" from their style.

All output from Lyria 3 and Lyria 3 Pro is embedded with SynthID, Google's invisible watermarking system for AI-generated content, which allows the audio to be identified as machine-made even after post-processing.

The launch lands amid wider industry moves on AI music attribution. Spotify this week began rolling out tools to let artists flag AI-generated music uploaded under their names, and Deezer earlier this year opened its AI detection tools to rival platforms.