Lyria 3 Pro is Google DeepMind's flagship music generation model, released in March 2026 as the premier tier of the Lyria 3 family. It represents a significant evolution in AI-assisted music production, transitioning from the short-form loops of earlier versions to full-length song generation. The model is designed to produce high-fidelity, production-ready audio that maintains thematic and melodic consistency across extended durations.
Structural Control and Composition
The defining feature of Lyria 3 Pro is its professional-grade structural awareness. Unlike standard generative models that produce audio based on a single continuous mood, Lyria 3 Pro allows creators to architect songs with specific sections, including intros, verses, choruses, bridges, and outros. This allows for precise control over song arrangements, enabling the model to manage transitions and dynamic shifts that mirror human composition techniques. It can generate complete tracks approximately three minutes in length.
Multimodal Capabilities and Audio Fidelity
The model produces 48kHz stereo audio and supports a wide range of musical genres and languages. It is a multimodal system that accepts both natural language text and image inputs; for example, an image can be used to influence the atmospheric "vibe" or aesthetic direction of a track. It supports realistic vocal generation with nuanced phrasing and expressive dynamics, alongside complex instrumentation and layered arrangements.
Safety and Transparency
To address safety and intellectual property concerns, Lyria 3 Pro integrates SynthID watermarking technology. This embeds an imperceptible digital signature directly into the audio waveform, allowing the content to be identified as AI-generated even after compression or editing. Additionally, the model utilizes specialized filters designed to prevent the direct mimicry of specific artists and ensures that outputs align with safety guidelines regarding copyrighted material.