Google launched Lyria 3.5, the latest version of its AI music-generation model, on Wednesday, July 29, rolling it out inside Google Flow Music, the company's AI song-creation platform Music Business Worldwide.

Google DeepMind said the upgrade delivers significant advancements across musicality, lyrics, and vocal quality. The model produces more realistic and emotionally nuanced vocals with improved pronunciation, supports richer and more complex melodic structures, and offers higher-quality lyrics with better adherence to user prompts. Users also get easier control over tempo and duration within Flow Music Music Business Worldwide.

The Lyria 3.5 product page adds that the model supports variable song lengths, lets users explore new genres and soundscapes, and is designed for real-time, interactive use Google DeepMind. Google also maintains a separate product, Lyria RealTime, built specifically for real-time interactive music generation and performance Google DeepMind.

The launch extends a product line that has moved quickly. Google DeepMind first unveiled the Lyria model in November 2023, in partnership with YouTube, alongside a YouTube Shorts experiment called Dream Track Music Business Worldwide. The original Lyria 3, launched in February 2026, was limited to generating 30-second clips; a month later, Google released Lyria 3 Pro, which can generate tracks up to three minutes long TechCrunch. Lyria 3 generates 44.1 kHz stereo audio — standard CD-quality resolution — and can generate music from images as well as text prompts Google AI. Google announced the Flow Music app at Google I/O 2026 Android Central.

When Lyria 3 launched in February, Google described it as "designed for original expression, not for mimicking existing artists" and said it had been "very mindful of copyright and partner agreements" Music Business Worldwide.

The company has never specified what music was used to train the Lyria model line. The Lyria 3.5 announcement did not disclose training data, made no mention of SynthID — Google's audio-watermarking technology — and carried no named author Music Business Worldwide.

That omission sits against an active lawsuit. In March, a group of independent musicians and songwriters sued Google, accusing it of training Lyria 3 on copyrighted recordings pulled from YouTube without permission or payment. The plaintiffs' complaint claims Google copied at least 44 million clips and 280,000 hours of music to train the model Music Business Worldwide.

Google moved to dismiss the lawsuit in June. The company's argument: each plaintiff granted YouTube and its affiliates a "worldwide, non-exclusive, royalty-free, sublicensable and transferable license" to use their uploaded content when they uploaded it to the platform Music Business Worldwide. The case is ongoing.

For the music industry, the pattern is now familiar. AI models get more capable with each release, and the legal questions about how they were built move more slowly than the technology itself. Google's Lyria line has gone from 30-second clips to variable-length, real-time-interactive song generation in under a year. Whether the training data behind it was lawfully obtained remains a matter for the courts.