Google has announced the integration of Lyria 3, the latest generative AI model for music developed by DeepMind, into the app Gemini. The feature, currently in beta, allows users to create 30-second songs from text prompts or by uploading images and videos.
The system processes the provided content to generate complete tracks with lyrics, vocals, and instrumental backing, introducing significant improvements in automatic lyric generation and control over parameters such as musical style, vocal timbre, and tempo. Lyria 3 also increases harmonic complexity and sonic realism compared to previous versions of the technology.
The tracks produced are accompanied by a custom cover generated by the Nano Banana model and include SynthID, an imperceptible digital watermark developed by Google to identify content created through artificial intelligence. In parallel, Gemini Updates its verification capabilities, allowing users to upload audio files to verify their origin using proprietary recognition systems. The model has been trained to favor original expression, limiting direct imitation of existing artists. If a specific musician is mentioned in the prompt, the system creates a song inspired by their general style or vibe, without replicating their identity.
The technology is also being implemented on YouTube for the creation of soundtracks for the Shorts. The feature is available to users 18 and older on desktop starting today, Wednesday, February 18, 2026, at 4:18 PM Swiss time, with a gradual rollout to mobile devices expected in the coming days. The service supports several languages, including German, French, Spanish, and English, with higher usage limits for subscribers to the Google AI Plus, Pro, and Ultra plans.