Google announced on Wednesday the addition of a music-generation feature to the Gemini application, powered by DeepMind’s Lyria 3 model. The feature is currently in beta and allows users to create songs by describing them. The tool generates a 30-second track containing lyrics and cover art. Users can also upload photos or videos to generate songs matching the media’s mood. The feature supports English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese. It is available globally to all users aged 18 and older.
The Lyria 3 model serves as the engine for this new capability. Google states that this model improves upon previous generations by producing more realistic and complex musical outputs. Users can control specific elements of the generated tracks, including style, vocals, and tempo. The generation process involves a user describing a song concept, such as a “comical R&B slow jam about a sock finding its match.” The application then processes this prompt to create a 30-second audio file accompanied by visual artwork, which is generated using a tool referred to as Nano Banana.
Google has integrated media upload capabilities into the music generation tool. Users can submit a photo or video file to the Gemini app. The AI analyzes the media to determine the mood. Based on this analysis, the tool generates a song designed to match the emotional tone of the uploaded file. This functionality expands the input methods beyond text-based prompts, allowing for visual inspiration to drive the musical creation process.
While users can include artist names in their prompts, Google has clarified the limitations of this functionality. The system does not replicate the voice or specific sound of a named artist. Instead, the model uses the name as a source of broad stylistic inspiration. The company stated that Lyria 3 is designed for original expression rather than mimicking existing artists. A blog post from the company explained, “If your prompt names a specific artist, Gemini will take this as broad creative inspiration and create a track that shares a similar style or mood.” To enforce this, filters are in place to check outputs against existing content.
Google is expanding the availability of Lyria 3 beyond the Gemini app. The model is being rolled out to YouTube creators globally through the Dream Track feature. Previously, Dream Track was restricted to YouTube creators in the United States. This update extends access to creators worldwide, allowing them to utilize the AI tool for generating tracks directly on the YouTube platform.
Security and identification of AI-generated content are central to the release. All songs created with the Lyria 3 model include a SynthID watermark. This watermark identifies the content as AI-generated. Additionally, Google is enhancing the Gemini app’s capabilities to detect AI music. Users will be able to upload audio tracks and ask Gemini to determine if the music is AI-generated. This detection feature utilizes SynthID to identify the presence of AI elements in uploaded files.
The release occurs within a complex landscape for AI-generated music. Companies like YouTube and Spotify are embracing the technology, signing contracts with music labels to monetize AI music. Conversely, AI model companies face ongoing copyright lawsuits from the music industry regarding the training data used for these models. Platforms such as Deezer have implemented tools to mark AI-generated music to prevent fraudulent streaming. Google’s introduction of SynthID watermarking and detection tools addresses concerns regarding the identification of synthetic media.
Featured image credit
