On February 18, 2026, Google unveiled that its chatbot application, Gemini, has now incorporated the latest audio model, Lyria 3, developed by DeepMind. This integration enables users to effortlessly generate AI-composed music tracks of approximately 30 seconds directly within the chat interface. Currently available as a beta version, this innovative feature facilitates music creation through diverse input methods, including text, images, and videos, eliminating the hassle of switching between different tools or platforms.

Initially launched in the United States, the feature is being gradually rolled out to YouTube creators globally. It supports multiple languages, such as English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese, ensuring accessibility to a wide range of users. Users have the flexibility to tailor music by specifying genre, mood, or scenario. For instance, by entering a prompt like "An Afrobeat song for my mom, with lyrics about the beautiful memories of growing up together," the system will automatically craft a track in the requested style. The generated output includes both instrumental pieces and songs with lyrics that are both composed and sung by AI.

Additionally, users can upload photos or videos, prompting Gemini to create music with lyrics that resonate with the mood and theme of the visuals. Google has also integrated Lyria 3 into YouTube's Dream Track feature, empowering creators to generate personalized AI background music for their Shorts videos.

To maintain transparency regarding the origin of the generated content, all tracks are discreetly watermarked using SynthID technology. This allows for the detection of whether the music was created or edited by AI. Google has underscored its meticulous approach to copyright concerns during the development of Lyria 3. When users explicitly mention a specific singer, Gemini refrains from directly replicating their vocal style. Instead, it generates tracks with similar styles or emotions and employs a filtering mechanism to compare the results with existing content, thereby mitigating potential infringement risks.