Google launches Lyria 3 for 30-second AI music generation in Gemini

Google launches Lyria 3 music creator

Google released Lyria 3 on Wednesday to let Gemini users generate 30-second songs using text prompts, photos, or video clips. The new model represents the latest iteration of Google DeepMind’s music generation technology and is now integrated directly into the Gemini Tools menu. This update moves the technology out of limited experimental phases and into a primary interface for general consumers.

Users can access the tool by selecting it alongside other AI generators like the Veo video maker or the Nano Banana image engine. The system handles the entire creative process from a single input, producing a fully arranged audio track with vocals and instrumentation. Google designed the interface to lower the barrier for entry for people who lack musical training or songwriting experience.

The company showcased the model’s capabilities with several specific examples during the announcement. One track featured an R&B arrangement focused on a narrative about lost socks in a washing machine. Another example utilized Afrobeat rhythms to describe childhood memories of cooking plantains with a parent. Both tracks showcased the model's ability to sync thematic lyrics with specific musical genres without human intervention.

Automated lyrics and visual art

Lyria 3 introduces a significant update to Google’s creative suite by generating original lyrics based on user prompts. Previous versions often required users to provide their own text or relied on instrumental outputs. Senior product managers Joël Yawili and Myriam Hamed Torres stated that the goal is to provide a fun way for users to express themselves. The AI writes the verses and choruses on the fly to match the requested mood or story.

The tool also automates the visual side of music sharing by creating custom cover art. Google uses its Nano Banana image generator to produce a square graphic that accompanies every generated audio file. This removes the need for users to upload their own sentimental photos or design their own branding for the 30-second clips. The result is a packaged digital asset ready for social media sharing or personal messaging.

Critics of generative AI note that these "creative" outputs come with a high environmental cost due to the massive compute power required to run the models. While Google pitches the tool as a way to say "I love you" through song, the underlying infrastructure relies on energy-intensive data centers. The company continues to balance these sustainability concerns with its aggressive push to integrate AI into every facet of the Gemini ecosystem.

Evolution from YouTube Dream Track

Google first introduced the Lyria brand in 2023 as part of a series of controlled experiments. Its most visible application was Dream Track, a feature inside YouTube Shorts that allowed a small group of creators to generate music in the style of famous pop stars. That initial version relied on direct partnerships with artists who agreed to let their voices and styles be synthesized by the AI.

The technology was also tested in the Music AI Sandbox, a more technical environment designed for professional musicians. That suite allowed artists to iterate on specific loops or stems rather than generating a finished 30-second pop song. Lyria 3 takes the learnings from those professional tools and simplifies them for the Gemini user base. It offers more control over vocals and musical complexity than the original Lyria model.

The current version of the engine remains limited to 30-second durations. While this fits the format of vertical video platforms like TikTok or YouTube Shorts, it is not yet clear when Google will expand the length for full-song compositions. The company is currently updating the Dream Track feature to use the Lyria 3 backend, though the 30-second cap remains the standard for now.

Copyright and artist training data

Google has not disclosed the full list of artists used to train Lyria 3, but previous iterations were built on data from T-Pain, Demi Lovato, and Sia. The company claims the new model is programmed to be mindful of copyright and existing intellectual property agreements. Rather than mimicking specific voices, the AI is designed for "original expression" that avoids direct imitation of famous performers.

If a user enters a prompt naming a specific artist, the system will treat that name as a broad stylistic suggestion. It might produce a track with a similar tempo or instrumentation, but it will not attempt to clone the artist’s specific vocal timbre. Google implemented filters to prevent the AI from generating melodies that are too similar to existing copyrighted songs. These filters act as a digital gatekeeper to reduce the risk of legal challenges from music labels.

Despite these safeguards, Google admits the system is not perfect and relies on user feedback to improve. The company has asked users to report any content that appears to infringe on copyright. This approach places some of the policing responsibility on the community as the AI continues to scrape and synthesize musical patterns. The legal landscape for AI music remains volatile as record labels continue to sue other startups like Suno and Udio for copyright infringement.

Global rollout and language support

The global rollout for Lyria 3 begins on February 18 for all Gemini users who are 18 years or older. Google is targeting a broad international audience by launching with support for multiple languages. The tool will be available on desktop browsers first, with a mobile app update following shortly after the initial release.

The initial launch includes support for the following languages:

• English
• German
• Spanish
• French
• Hindi
• Japanese
• Korean
• Portuguese

Google plans to add more languages and regional styles as the rollout continues throughout the year. The staggered release ensures that the company can monitor server loads and copyright filter performance in real-time. While some users may see the tool in their Gemini sidebar immediately, others will receive access over the next several days as the update propagates across global servers.

This move puts Google in direct competition with specialized music AI startups. By embedding Lyria 3 directly into Gemini, Google is betting that convenience will win out over the more specialized features found in standalone music apps. The integration allows for a seamless workflow where a user can ask Gemini to write a poem, turn it into a song, and generate a video—all within the same chat interface.