Lyria 2 is a generative AI model developed by Google DeepMind for high-fidelity music and audio production. It is designed to produce professional-grade 48kHz stereo audio across various musical genres and complex compositions. The model is integrated into Google’s Music AI Sandbox and is accessible via the Vertex AI platform.
The model offers granular creative controls, allowing users to specify parameters such as BPM, musical key, and instrumentation through text prompts. Key features include the ability to generate singing vocals from lyrics, extend existing tracks, and perform musical "inpainting" to bridge audio segments. To ensure safety and provenance, Lyria 2 incorporates SynthID watermarking, which embeds an imperceptible digital signature directly into the generated audio.