A Musical breakthrough in Gemini: Google taught its AI to create songs from a single phrase
Google DeepMind has integrated a new music model, Lyria 3, into its Gemini chatbot. The neural network can now generate full 30-second tracks complete with vocals, lyrics, and even cover art directly within the chat window — without any external plugins or services.
The key advantage of Lyria 3 is full autonomy. Users no longer need to write song lyrics themselves: simply describing a mood or situation is enough. The AI composes the lyrics, selects the melody, and creates the arrangement on its own. Users can now control tempo, style, and even the type of vocals, resulting in a more professional and immersive sound.
The tool can also “read” the atmosphere of your media files. Upload a photo of a dog on a mountain trail, and Gemini may suggest a folk or ambient-style track. The visual component is handled by a neural network called Nano Banana, which automatically generates cover art for each composition.
Google has implemented copyright safeguards: the AI does not copy the voices of real stars but instead uses their style as a reference. Each file contains an invisible SynthID watermark. It is inaudible but allows users to instantly verify through Gemini whether a track was created by artificial intelligence.
The feature is already available in the web version of Gemini for users over 18, with mobile app support coming soon. The tool supports eight languages and is open to YouTube Shorts creators worldwide. Subscribers to the Plus, Pro, and Ultra plans have received expanded limits on the number of tracks they can generate.
ORIENT
