Google Lyria 3 Pro: AI Music Generator Now Creates 3-Minute Songs
Google today announced Lyria 3 Pro, an advanced iteration of its AI music generator, now capable of creating tracks up to three minutes long. The upgrade, available through the Gemini app, expands on the initial Lyria 3 release from last month, which limited song lengths to 30 seconds.
Lyria 3 Pro offers improved understanding of musical composition, allowing users to prompt for specific song elements such as intros, verses, choruses, and bridges, according to Google. The new version is aimed at users seeking greater customization and creative control over their AI-generated music.
Access to Lyria 3 Pro is tiered based on Google AI subscription levels. Subscribers to Google AI Plus receive 10 tracks per day, Pro subscribers receive 20, and Ultra subscribers receive 50, according to Google. The company suggests use cases including custom soundtracks for video blogs, podcasts, or instructional videos.
Beyond the Gemini app, Lyria 3 and Lyria 3 Pro are also being integrated into Google Vids, enabling users to add custom music to their video projects. This functionality will be available to Google Workspace customers and Google AI Pro/Ultra subscribers starting this week.
For developers and enterprise users, Lyria 3 Pro is now available in public preview on Vertex AI, allowing for scalable, on-demand audio generation. Google states this will enable organizations to create custom game soundtracks or integrate the technology into creative tools, music, and video platforms. The model has also been enhanced within Google AI Studio and the Gemini API, providing developers with greater musical awareness and structural consistency for building next-generation creative tools.
ProducerAI, a collaborative music creation tool launched recently, is also leveraging Lyria 3 Pro to offer an agent experience designed to assist artists, producers, and songwriters of all skill levels in composing full-fledged tracks. ProducerAI is available to both free and paid subscribers globally.
Google emphasized safety and responsibility in the development of Lyria 3, stating the model was trained using materials from YouTube and Google that the company has the rights to use under its terms of service, partner agreements, and applicable laws. The model is designed to avoid imitating artists; if a creator is specified in a prompt, it is treated as broad inspiration. Output is also checked against existing content through filters, and all generations include a SynthID watermark.
