Google DeepMind released Lyria 3 Pro, pushing the boundary of generative music from a few seconds of clip directly to a full 3-minute track. This is not simply "extending the generation time" — the core lies in the model's explicit understanding of musical structure: it can now prompt out traditional composition elements like intro, verse, chorus, and bridge, and control style switches and complex transitions, essentially upgrading generation from "sampling-based splicing" to "composition with structural awareness." Structured output means the model has learned "how a song is written," not just "how a piece of audio continues."
More noteworthy is Google's full-stack landing strategy: Vertex AI gives enterprises on-demand audio production, AI Studio and the Gemini API let developers access real-time audio streams, Google Vids gives ordinary creators one-click scoring, Gemini App gives consumers personalized tracks, and ProducerAI introduces an "agentic" experience — turning Lyria 3 Pro into a music producer that can continuously collaborate, with artists iterating complete works section by section. Combined with Lyria RealTime's streaming output, Google simultaneously occupies the "long structure," "real-time stream," and "multi-product matrix" three quadrants, upgrading AI music from "single-point tool" to "ecosystem covering the entire creative chain."
On the safety side, all outputs are embedded with SynthID watermarks, and the model is explicitly stated not to mimic named artists. Grammy producer Yung Spielburg has already used Lyria to write film scores, and DJ François K is iterating singles with Lyria. If the competitive point of Suno and Udio is still "audio quality of a few seconds," Lyria 3 Pro's differentiation lies in "structured, controllable, producible." From generating a clip to generating a song to agentic creation, the dimensions of this competition are being rewritten.