Lyria 3 Pro release: DeepMind extends music model to longer tracks

DeepMind has released Lyria 3 Pro, a music-generation model designed to produce longer pieces with structural awareness, and is bringing the model to more Google products and surfaces. The Pro variant focuses on preserving musical form across extended durations so generated tracks maintain recurring themes, transitions, and section-level coherence.

Lyria 3 Pro builds on prior Lyria models by extending the model's capacity to reason about structure over longer timeframes. Rather than stitching short motifs together, DeepMind says the Pro model tracks higher-level musical elements so sections like verse, chorus, bridge and outro can develop with internal consistency. The model supports user controls for style, instrumentation and high-level arrangement, and aims to reduce abrupt or repetitive loops that can undermine longer generated tracks.

What Lyria 3 Pro does

The headline capability is structural awareness at increased track length. Where earlier music models often generated short loops that required post-production to produce a full song, Lyria 3 Pro is positioned to generate output that sounds like a continuous composition with intentional changes and callbacks. DeepMind frames the improvement as both a qualitative advance in musical form and a usability improvement for creators who want longer, ready-to-use outputs.

DeepMind highlights two areas of improvement. First, the model maintains global coherence across sections, which helps with motifs, harmonic progression and rhythmic continuity. Second, it offers controls for higher-level composition choices, so a user can ask for specific section ordering or emphasize a recurring melodic idea. DeepMind emphasizes safety and rights considerations, noting the standard content filters and policies that apply to generative audio, and saying the model was trained to avoid directly reproducing copyrighted recordings.

Under the hood, DeepMind describes architectural and training enhancements meant to preserve context across a larger window of musical time. The company positions Lyria 3 Pro as the incremental, performance-focused sibling to Lyria 3, offering longer duration outputs with the same stylistic range. Exact technical specifications and parameter counts were not disclosed in the announcement.

Availability and product integration

DeepMind is rolling Lyria 3 Pro out to more Google surfaces, promising integrations beyond research demos and developer kits. The company said the model will appear across additional Google products and surfaces, making the capability available to a wider set of users and workflows. DeepMind did not provide a granular product-by-product rollout timeline in the announcement, instead saying availability will expand through Google channels over the coming months.

For creators and developers, DeepMind expects Lyria 3 Pro to be accessible via existing APIs and developer programs where Lyria models are supported. Licensing, usage limits and pricing details were not published with the initial release; developers should expect platform-specific terms once individual Google products integrate the model.

Why it matters

Longer, structurally aware music generation lowers the friction for producing end-to-end tracks, which affects composers, game developers, advertisers and content creators who need longer pieces without manual editing. Wider integration into Google products increases the model's reach, shifting Lyria from a research offering toward a practical tool in consumer and creative apps. The move also raises questions about copyright, quality control and the economics of music production as automated tools become capable of producing finished-length material.