Lyria 2
DeepMind's high-fidelity music model with real-time steerable generation
About
Lyria 2 was announced on April 24, 2025 as Google DeepMind's latest music generation model, aimed squarely at professional musicians rather than casual prompters. It powers the Music AI Sandbox's Create and Extend tools — describe a sound, genre, mood, or vocal style and it generates or continues tracks — plus Lyria RealTime, which lets performers steer music as it plays. Outputs carry SynthID watermarks. Distribution followed the usual Google playbook: YouTube's Dream Track experiment, a September 2025 Speech-to-Song feature, and eventually the Gemini API and Vertex AI. Caveats: initial access was waitlist-only for US musicians; it is a closed model with no weights; and the ethics of training data and artist consent remain live debates around it, as with all commercial music models.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported: professional-grade fidelity across genres and intricate compositions
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Announced April 24, 2025 by Google DeepMind; multiple outlets confirm date and Hassabis tweet; Powers Music AI Sandbox; YouTube Dream Track and Speech-to-Song integrations documented; Closed model, waitlist access; training-data ethics caveats noted.