Lyria 3 is the music generation model built into Gemini (description: https://deepmind.google/models/lyria/ and direct access: https://gemini.google.com/music).

The 30-second songs can be generated in several ways:

• Remix from homepage styles and an instruction
• text to music: an instruction with genre, rhythm, instruments, etc.
• image to music: a source image with an instruction
• Video to Music: A source video with an instruction

Lyrics can be added (excluding authors’ works, royalty-free included – see screenshot with Verlaine’s Poetic Art) by being “tagged” with the type “[1st verse], [Chorus]” and the songs are marked with Google’s SynthID watermark.

Generations come with a cover image generated by Nano Banana.

8 languages are available: English, French, German, Korean, Hindi, Japanese, Portuguese, Spanish.

The 3 models – fast, reasoning and pro – of Gemini 3 can be selected (fast by default).

Export mp4 with animation and subtitles or mp3.

Limitations related to the type of account Gemini used. No limits observed after 12 generations with a free account.

Remix

Text to music

Image to music:

Video to music (source: https://www.pexels.com/fr-fr/video/vue-aerienne-nocturne-des-rues-animees-de-la-ville-35925935/):