Lyria 3 Pro: Create longer tracks in more
read at source ↗ deepmind.google
Lyria 3 Pro: Create longer tracks in more
Source: DeepMind Date: 2026-03-25 URL: https://deepmind.google/blog/lyria-3-pro-create-longer-tracks-in-more/
Summary
Google released Lyria 3 Pro with structural music awareness enabling compositions up to 3 minutes with controllable song structure (intro, verse, chorus, bridge). Trained on YouTube/Google licensed material. Available across Vertex AI, AI Studio, Gemini API, Google Vids, Gemini app, and ProducerAI. SynthID watermarking on all outputs. No artist mimicry; copyright filtering against existing content. Tested with Grammy-winning producer Yung Spielburg.
Implications
3 minutes with controllable song structure is the professional threshold. The previous 30–90 second range was demo-quality; 3 minutes with intentional verse/chorus/bridge structure is the minimum for a usable song sketch. That’s the difference between “interesting toy” and “useful creative tool” for working musicians and music directors.
Six-platform distribution is the same playbook as Veo and Imagen. Vertex AI + AI Studio + Gemini API + Google Vids + Gemini app + ProducerAI covers enterprise, developer, consumer, and specialized music production surfaces simultaneously. Google is not positioning Lyria 3 Pro as a standalone product — it’s infrastructure embedded in existing workflows.
Licensed training data + copyright filtering is the legal moat. Lyria 3 Pro’s explicit training data provenance (YouTube/Google licensed material) and content filtering against existing works positions it more defensibly than Suno and Udio, which face ongoing legal challenges from major labels. For enterprise buyers with IP risk exposure, that licensing posture matters.
Watch:
- ProducerAI adoption — a dedicated music production platform integration signals Google is serious about professional music workflow penetration
- Whether the 3-minute limit expands and whether full-song-length generation follows
- Major label responses: does Google’s licensing approach lead to formal partnership agreements that further legitimize AI music generation?