Google DeepMind has models for text, images, videos and even world models, and it’s now made a big play in music.
The AI research lab has launched Lyria 3, its most advanced generative music model to date, rolling it out today in beta inside the Gemini app for users aged 18 and over across English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese.
What Lyria 3 Can Do
Lyria 3 allows users to generate 30-second music tracks from text prompts or uploaded images and videos — complete with custom lyrics, vocals, and cover art. A user could, for instance, describe a mood, a genre, or a personal memory and receive a fully produced track within seconds. The model gives creators more granular control than its predecessors, with the ability to steer tempo, vocal style, and lyrical content.
Three capabilities mark Lyria 3 as a step beyond earlier Lyria models: automatic lyric generation based on user prompts (no need to supply your own words), greater creative control over musical elements, and the ability to produce more realistic and musically complex outputs. A companion cover art feature powered by Nano Banana rounds out each generated track, making it easy to share directly via download or link.
Beyond consumer use, Lyria 3 is also being integrated into YouTube’s Dream Track feature for Shorts creators in the U.S., with a broader international rollout underway.
Safety and Copyright
Google has embedded SynthID — its imperceptible AI-content watermark — into every track generated through the Gemini app. The company has also extended SynthID verification to audio, meaning users can upload a file and ask Gemini to check whether it was produced by Google AI. On the copyright front, DeepMind says Lyria 3 is designed for original expression rather than artist imitation; prompts that name specific artists will be interpreted as broad stylistic inspiration rather than direct mimicry, and output filters are in place to check generations against existing content.
A Crowded and Fast-Moving Competitive Landscape
Google’s move into consumer music generation puts it squarely in competition with a rapidly expanding field. Suno and Udio have built dedicated platforms for AI music creation and have each attracted millions of users, though both face ongoing legal scrutiny from major record labels over training data practices. Meta has released its AudioCraft suite, including the MusicGen model, as an open-source offering. OpenAI has explored audio generation capabilities, while startups like Stability AI have also entered the space with their own music models.
Where Google has a structural advantage is distribution. By embedding Lyria 3 directly into the Gemini app — which already serves a large and growing user base — DeepMind bypasses the cold-start problem that standalone music AI platforms face. Tighter integration with YouTube, the world’s dominant music streaming and discovery platform, gives Google additional leverage that no pure-play music AI competitor can currently match.
The broader race to own generative media is intensifying: text generation is largely commoditized, image generation is maturing, and video generation is emerging as the next frontier. Music, until now, has been a conspicuous gap in Google DeepMind’s consumer-facing AI portfolio. Lyria 3 could help close it.