Lyria 3
2026-02-18 · Publisher : Google · Domain : Artificial Intelligence
Google DeepMind launches Lyria 3, its music generation model, in beta in the Gemini app: 30-second tracks with generated lyrics, from a text prompt, a photo or a video.
What changes
- Lyrics are generated automatically from the prompt, with no text supplied by the user.
- More control over style, vocals and tempo, and more realistic and complex tracks.
- Possible inputs: a text description, or a photo or video used as inspiration.
- Cover art generated by Nano Banana; output marked with SynthID, with audio verification added in Gemini.
- Also available in YouTube Dream Track for Shorts.
### Specifications - Output: 30-second tracks in the Gemini app. - Availability: Gemini app (beta) and YouTube Dream Track; API access via the Lyria suite (see pricing). - API identifier, context window and knowledge cutoff: not disclosed in the post. ### Pricing | Item | Pricing (current pricing page, not the one at announcement) | |---|---| | Lyria 3 Clip (preview) | USD 0.04 per track | | Lyria 3 Pro (preview) | USD 0.08 per track | ### Benchmarks No scores published in the announcement. ### Key takeaways A consumer music generation feature in beta, with SynthID marking; developer access arrived with Lyria 3 Pro the following month.
Links
Tags : Audio, Google, Lyria, Musique, Génératif