Google brings AI music generation to Gemini with new Lyria 3 model

Google announced on Wednesday that it has integrated its new Lyria 3 model into Gemini, letting users generate and remix 30-second music tracks from simple prompts. This expands Gemini’s existing multimodal capabilities and will also power YouTube’s “Dream Track” feature for Shorts.

Users will be able to generate music using basic text prompts, such as a “comical R&B slow jam”, or get highly granular by altering tempo, drumming styles, and emotional mood.

Gemini’s Lyria 3 goes way beyond text. It allows users to generate tracks based on uploaded photos or videos. Surprisingly, these tracks can even be paired with AI-generated album art powered by Google’s Nano Banana image model.

According to a Google blog post, Lyria 3 represents a substantial leap in producing realistic, musically complex tracks with professional-grade arrangements and automated lyric writing. However, the music generation is limited to 30-second clips; the model offers exceptional control over individual song components.

All Lyria 3 tracks utilize Google’s SynthID watermarking for transparency, allowing for identification with the SynthID Detector launched at Google I/O 2025. While the instrumental music is very realistic, initial feedback indicates the AI-generated lyrics can sometimes lack a natural sound.

Lyria 3 has rolled out worldwide for Gemini users aged 18 and older. Currently, it supports prompts in English, Spanish, German, French, Hindi, Japanese, Korean, and Portuguese.