Google's Lyria 3 Pro extends AI music from a jingle to an actual song

What Happened

The base Lyria 3 launched February 18 with 30-second tracks generated from text prompts. Thirty seconds is roughly the length of a TV spot. Three minutes is a song. The Pro tier does not just extend duration — it understands musical structure: intros, verses, choruses, bridges. A model that knows where a chorus belongs is doing something different from one that generates audio until it stops.

Lyria 3 Pro rolls out this week to Google AI Pro and Ultra subscribers and Workspace customers, with availability in Google Vids (video soundtracking), Vertex AI (enterprise-scale generation), AI Studio (developers), and ProducerAI, a platform built by working musicians. Every track carries SynthID watermarking for AI detection. Google developed the model with input from producers and musicians, and it supports vocal generation across eight languages.

Why It Matters

The product spread — across a consumer subscriber tier, enterprise cloud infrastructure, a developer API, and a musician-facing platform — signals that Google is positioning AI music as infrastructure, not a feature. Suno, its most direct competitor, raised $250 million in November 2025. Google's response is integration depth, not a standalone product.

The copyright question has not been resolved cleanly. Google asserts that Lyria 3's training data is fully licensed under terms-of-service and partner agreements. That claim matters because Billboard reported in January 2024 that earlier Google music AI models were trained on copyrighted recordings before licensing deals were in place. A Universal Music Group and YouTube licensing agreement from October 2025 created some formal structure, but there is no independent verification of Google's training-data assertions for Lyria 3 specifically. Google acknowledges its copyright filters may not catch everything.

Key Takeaways

• Lyria 3 Pro is the first version of Google's AI music model capable of producing something resembling a complete track. The gap between its stated capabilities and its unverified copyright disclosures is the story worth watching.