Google DeepMind has introduced the Lyria 3 AI model for music generation. The corporation has integrated it into the Gemini chatbot.
The neural network enables audio generation using text prompts, photos, or videos.
“Simply describe an idea or upload an image, for example: ‘a comical slow R&B track about a sock that found its pair,’ and within seconds, Gemini will transform it into a high-quality and memorable composition,” the announcement states.
Compared to previous versions, Lyria 3 has been “enhanced” in three areas:
- there is no need to write your own text — the LLM will create it based on the prompt;
- creative control over style, vocals, and tempo;
- the ability to create realistic and musically complex tracks.
Gemini creates 30-second audio with a custom cover generated by Nano Banana. The result can be shared with friends.
“The goal is not to create a musical masterpiece, but to provide a fun and unique way of self-expression,” noted Google.
The model is available in English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese. Initially launched on desktop, the mobile version will be available in the coming days. Subscribers to Google AI Plus, Pro, and Ultra will receive extended limits.
AI Labeling
AI-generated music is becoming increasingly common. Streaming service Deezer has 9.7 million paying users. It recorded over 50,000 AI tracks uploaded daily — about a third of the total.
Meanwhile, 97% of listeners cannot distinguish AI-generated songs from those written by humans.
All tracks from Lyria 3 receive an embedded SynthID label — an invisible watermark that allows for AI content identification.
Back in September 2025, Suno introduced the “world’s first” generative DAW — Suno Studio, which “radically reimagines the music creation process.”