Google DeepMind rolled out Lyria 3, its newest generative music model, inside the Gemini app's beta version. Users can type a prompt or upload an image, and within seconds the system composes a 30‑second track with lyrics, vocals and custom cover art. The tool aims to give everyday creators a quick, playful way to add original soundtracks to messages and social posts.

Beyond novelty, Gemini embeds each piece with SynthID, an invisible watermark that flags AI‑generated audio. Users can ask the app to scan any file and receive a verdict, extending the same verification used for images and video. Lyria 3 currently supports English, German, Spanish, French, Hindi, Japanese, Korean and Portuguese, with desktop rollout today and mobile arrival in the coming days.

Creators on YouTube Shorts can already tap Lyria 3 via the Dream Track feature, where the model supplies richer verses or backing beds tailored to each clip. While DeepMind stresses the service is for original expression—not imitation of specific artists—filters block direct copying and users may report violations. Availability today for anyone 18+ makes custom jingles as easy as sending a meme.