Google DeepMind has shipped Lyria 3.5, its latest music generation model, now live inside Google Flow Music. The update targets four specific weaknesses in prior versions: musicality, lyric coherence, vocal quality, and user-level creative control. These are not minor refinements. They represent the core failure points that made earlier AI music tools feel like demos rather than production instruments.
What makes this worth reading beyond the headline is the detail on creative control. The model gives users levers over musical structure that previous versions abstracted away. That shift, from black-box generation to steerable output, is where the real engineering story lives. The vocal improvements are also technically significant, addressing pitch consistency and phoneme alignment problems that have plagued text-to-audio pipelines across the industry.
The immediate question is whether Lyria 3.5 closes the gap between AI-assisted music and professional production workflows, or whether it remains a tool for ideation only. Google Flow Music is the delivery vehicle, but the model architecture is what matters long-term. Read the full DeepMind post for the technical specifics on how each of the four capability areas was actually improved.
[READ ORIGINAL →]