Google DeepMind launches Lyria 3.5 in Flow Music

Google DeepMind released Lyria 3.5, the newest music-generation model powering Google Flow Music. The update spans four dimensions the team highlighted: musicality, lyrics, vocals and creative control, with vocals specifically described as more expressive and dynamic. It continues Google's steady cadence of generative-media models alongside its language-model releases this week.

The strategic context is Google's breadth play: the same week it shipped Gemini 3.6 Flash, expanded managed agents, began Gemini 4 pretraining and pushed Gemini toward a billion users, DeepMind advanced generative music. This full-spectrum coverage — text, agents, music, and (via the AlphaFold reshuffle) science — is Google leveraging its research depth to compete on every generative front simultaneously.

Competitive context: AI music generation is a contested space with Suno, Udio and others facing significant copyright litigation from rights holders. Google's advantage is distribution (Flow Music, YouTube adjacency) and its ability to negotiate licensing at scale; the risk is the same legal exposure around training data that has dogged the category. More expressive vocals in particular raise voice-likeness and consent questions.

Caveats and what to watch: DeepMind's blog announcement is light on benchmarks or technical detail, and 'advances across musicality' is subjective without comparative samples. The meaningful questions are licensing terms, how tightly Lyria 3.5 integrates into consumer YouTube/Flow surfaces, and whether Google can differentiate on quality and rights-cleared training data in a category increasingly defined by legal risk rather than model capability.