Google has launched Lyria 3, the latest generative music model developed by Google DeepMind, making it available in Arabic (beta) worldwide. The tool is being rolled out on the Google Gemini website, with mobile access expected to follow in the coming days.
With Lyria 3, users can generate 30-second custom music tracks by simply describing what they want to hear. For example, a user can type, “an upbeat, modern Arabic fusion track for Ramadan”, and Gemini will create a high-quality audio track within seconds. The feature is available to all Gemini users.
The model can be activated in three formats. Under the “Text to tracks” option, users can describe a genre, theme or mood to create original tracks with lyrics or instrumental audio. An example prompt is: “Create a track about my favorite meal my mom used to make, it was made of rice and beans. Surprise me!”
Another feature allows users to generate music from visual content. By uploading a photo or video, Gemini composes a track with lyrics that match the mood of the content. For instance: “Use these photos to create a track about my cat Katkoot on a hike in the mountains.”
Each 30-second track created in the Gemini app comes with custom cover art generated by Nano Banana, allowing users to download or share the track easily through a link. Google said the objective is not to produce professional-level music, but to provide a creative and personal way for users to express themselves.
All tracks generated using the Gemini app are embedded with SynthID, Google’s imperceptible watermark designed to identify AI-generated content. In addition, users can upload audio files to check whether they were created using Google AI. Gemini will scan for SynthID and apply its own reasoning before providing a response.
Google said music generation through Lyria 3 is intended for original expression rather than imitation of existing artists. If a user mentions a specific artist in a prompt, Gemini will interpret it as general creative inspiration and produce a track reflecting a similar style or mood. The platform also applies filters to ensure outputs do not replicate existing copyrighted material.
