Google Launches Lyria 3 for AI Music Generation

- All audio generated by Lyria 3 is embedded with SynthID, an imperceptible watermark used to identify the content as AI-generated; the Gemini app has also been updated to allow users to upload an audio file to check if it contains this watermark. - The model generates 30-second stereo tracks at a 48kHz sample rate in 16-bit PCM format and can be prompted by text, images, or video. - Lyria 3 represents a significant step from its predecessors; Lyria 1 (December 2023) was an early experiment for YouTube, and Lyria 2 (April 2025) was an instrumental-only API on Vertex AI, while the latest version automatically generates lyrics and is integrated into a major consumer application. - A key architectural feature is the Lyria RealTime API, which uses a chunk-based autoregression system to generate audio in 2-second segments, enabling real-time steering and control which is crucial for interactive applications on edge devices. - To prevent misuse, Google states the model is designed for original expression and not for mimicking specific artists; if a prompt names an artist, it is used for broad inspiration, and filters are in place to check outputs against existing content. - The model's integration into YouTube's Dream Track feature is expanding beyond its initial U.S.-only availability, allowing Shorts creators in more countries to generate soundtracks. - While Lyria 3 focuses on 30-second clips, it faces competition from startups like Suno and Udio, which have gained traction by allowing users to create full-length songs from text prompts. - The feature is rolling out to users 18 and over in eight languages, with free users having access and subscribers to Google AI Plus, Pro, and Ultra receiving higher usage limits.