Google Launches Lyria 3 Pro Music Generation Model, Capable of Creating Tracks Up to 3 Minutes Long

Google has significantly upgraded its AI music generation capabilities, launching the Lyria 3 Pro music generation model, which can create tracks up to 3 minutes long.

On Wednesday, March 25th, Google announced the launch of the Lyria 3 Pro music generation model, just one month after the release of the previous generation, Lyria 3.

The new model significantly increases the maximum track duration for a single generation from 30 seconds to 3 minutes, while also supporting fine-grained customization of musical structure elements.

Lyria 3 Pro will be simultaneously available on Gemini (paid version only), the video creation tool Google Vids, the AI music production platform ProducerAI, and enterprise solutions like Vertex AI, Gemini API, and AI Studio, covering multiple core scenarios from consumer to enterprise.

This series of deployments indicates that Google is accelerating the integration of AI music capabilities into its product ecosystem and seeking to open up new revenue streams in the enterprise market.

Core Upgrade, From Snippets to Complete Tracks

Compared to Lyria 3, the most significant changes in Lyria 3 Pro lie in its generation duration and structural awareness.

The new model supports the creation of tracks up to 3 minutes long and can recognize and generate specific musical structure elements, including intros, verses, choruses, and transitions. This means users can directly specify the complete song architecture in their prompts, rather than just describing the style or mood.

Google stated that Lyria 3 Pro is particularly suitable for creative scenarios requiring complex transitions or experimentation with multiple styles, providing a more complete solution for content creation such as podcast scoring, Vlog background music, and tutorial videos.

Google is attempting to enhance the credibility of the Lyria product series within the creative community through collaboration with professional musicians.

Grammy Award-winning producer Yung Spielburg has already used Lyria for soundtrack creation and production processes in Google DeepMind's short films.

DJ and producer François K has used Lyria iteratively to create a new song set to be released.

Google's Music AI Sandbox project also continues to provide experimental tools for musicians, producers, and songwriters, with feedback from these users being incorporated into the development of Lyria 3.

Multi-Platform Deployment, Covering Developers, Enterprises, and Creators

Google is extending the reach of Lyria 3 Pro to multiple different audience groups this time.

For enterprise users, Lyria 3 Pro is now in public preview on Vertex AI, supporting on-demand, large-scale generation of high-fidelity audio for applications such as rapid customization of game soundtracks and integration with creative tools.

For developers, Lyria 3 Pro is available on AI Studio, running in parallel with Lyria RealTime, and can be accessed via the Gemini API for developers to build next-generation creative tools.

For general users, Google's AI video creation app Google Vids is rolling out Lyria 3 and Lyria 3 Pro features to Google Workspace customers and Google AI Pro and Ultra subscribers starting this week, allowing users to add custom soundtracks to their creative projects or marketing videos.

The long-track generation feature in the Gemini app is also concurrently available to paid users.

Regarding ProducerAI, Google completed the acquisition of this AI music production tool last month. Following the integration of Lyria 3 Pro, ProducerAI will offer an intelligent, agent-like creation experience for artists, producers, and songwriters, and is now available to free and paid users globally.

Copyright and Provenance, SynthID Tagging and Training Data Explanation

Amidst ongoing industry concerns about the copyright of AI-generated content, Google has disclosed the model's data sources and content identification mechanisms.

Google stated that Lyria 3 Pro's training data comes from partner-licensed data and publicly available data from YouTube and Google platforms.

The company emphasized that the model does not directly imitate the style of specific artists, but if a user specifies an artist in their prompt, the model will draw "broad inspiration" from that artist's style to generate the track.

All audio tracks generated by Lyria 3 and Lyria 3 Pro will be embedded with SynthID watermarks to identify their AI-generated attributes.

This move aligns with overall industry trends. Spotify reportedly launched a new tool this week allowing artists to audit songs released under their name to prevent misattribution of AI-generated content. Streaming platform Deezer has also introduced a tool that any streaming service can use to identify AI music.