Google rolled out Lyria 3.5, the newest version of its music-generation model, on July 29 inside Flow Music, its song-making platform. Google DeepMind pitched it as a step up in musicality, lyrics and singing, with vocals that carry more feeling and clearer pronunciation, richer melodies, lyrics that follow prompts more closely, and easier control of tempo and track length. A demo song in the style of a female pop singer accompanied the launch. The model is available inside Flow Music rather than as a separate product.
What stood out to Music Business Worldwide was what the post left out. It did not explain how the model was trained, did not mention SynthID, the watermark Google adds to its AI audio, and dropped the copyright reassurances that came with Lyria 3 in February. Back then Google said that model was built for original expression rather than imitating artists, and that it had been careful about copyright and partner agreements while working with the music community. The new post had no named author either, where the Lyria 3 launch had carried the bylines of two product managers.
Google has never listed the music behind the Lyria line. Its standing line is that it trains on material it and YouTube have the right to use under their terms of service, partner agreements and the law.
The omission matters because Google's training is being challenged. In March a group of independent musicians and songwriters sued, saying Lyria 3 was trained on tens of millions of YouTube clips without consent or payment, and that owning YouTube and Content ID gives Google unusual leverage. Google's June motion to dismiss leans less on fair use than on YouTube's terms, arguing uploaders granted it and its affiliates a broad, royalty-free licence, and calling the claim that it trained on their specific songs unsupported.
Lyria dates to November 2023 and YouTube's Dream Track experiment, which cloned the voices of artists such as Charlie Puth and T-Pain for Shorts. Flow Music began as Riffusion, became ProducerAI, and was bought by Google in February; Believe now offers it to its artists, and recent additions like stem splitting bring it closer to Suno and Udio, both still in litigation with some of the majors.
Sources · 1