The underlying Lyria 3 model, developed by Google DeepMind, produces 30-second, 48-kHz stereo tracks from text prompts or uploaded photos and videos. Users can specify a genre, mood, or personal memory, and Gemini generates a complete track with vocals, lyrics, and a cover image. Compared with earlier Lyria versions, Lyria 3 writes song lyrics autonomously, offers greater control over style, voice, and tempo, and delivers more musically complex results.

The feature is available starting today on desktop and will roll out to the mobile app in the coming days. It is accessible to users aged 18 and over in eight languages, including German and English. Subscribers to Google AI Plus, Pro, and Ultra receive higher usage limits. If a user names a specific artist in a prompt, Gemini is designed to use the artist only as a broad stylistic inspiration rather than directly imitating them. According to Google, filters check generated outputs against existing content, and users can report potential copyright violations.

On YouTube, Lyria 3 is available via Dream Track, initially in the United States, with a rollout to additional countries underway. Google has not specified exactly where the feature is available within the Gemini app. All generated tracks include a SynthID watermark to indicate AI-generated content.

Rapid advances in generative AI for music—along with early streaming chart successes and lawsuits from Sony Music and other labels—show that AI-generated music has moved well beyond a novelty. While Google emphasizes that the feature is primarily “for fun” and says it handled copyright and partner agreements “carefully” during Lyria 3’s training, it has not disclosed detailed information about training data or commercial arrangements.

OpenAI is also reportedly working on a music generator for ChatGPT. At present, the only other widely regarded high-quality AI music generator is Suno; competitor Udio was acquired by Universal late last year and is currently inactive.