blog.google web signal

Google Lyria 3 Pro Rolls Out With 3-Minute Tracks

google music music-generation product-update

Key insights

  • Lyria 3 Pro now generates tracks up to three minutes with explicit verse, chorus, and bridge section prompting.
  • All Lyria 3 Pro outputs carry SynthID watermarks and the model refuses to imitate named artists by design.
  • Stability AI's Stable Audio 3 launched the same week with open-weight models supporting outputs up to 6 minutes 20 seconds.

Why this matters

Google deploying Lyria 3 Pro across six products at once tests whether a tightly integrated, watermarked, API-only music model can compete with open-weight alternatives that developers can self-host and fine-tune without usage restrictions. The SynthID watermarking layer is the first at-scale implementation of audio provenance tracking in a commercially deployed music model, setting a precedent that regulators and rights holders will watch closely as AI music litigation accelerates. For founders building in the audio or creator tooling space, the verse-chorus structured prompting interface represents a UX floor that any competing product now needs to clear to justify differentiation.

Summary

Google's Lyria 3 Pro music generation model is now live across six products simultaneously: Vertex AI, Google AI Studio, the Gemini API, Google Vids, the Gemini app, and ProducerAI. The rollout adds tracks up to three minutes long and structured prompting that lets users specify song sections like intros, verses, choruses, and bridges rather than just describing a vibe. Every output is watermarked via SynthID, Google's steganographic tagging system, and the model is built to decline requests that explicitly invoke named artists. That combination of reach and guardrails reflects how seriously Google is treating both enterprise adoption and copyright exposure. Essentially: Google is pushing Lyria 3 Pro into every surface it owns while Stability AI's Stable Audio 3 landed the same week with open-weight models and longer 6-minute 20-second outputs. - Structured verse-chorus prompting is a meaningful step beyond text-to-music -- it shifts control closer to actual songwriting workflow - SynthID watermarking is embedded invisibly and survives common audio processing, giving Google a provenance story competitors lack - Stable Audio 3's open-weight approach means developers can run it locally, a distribution advantage Lyria 3 Pro's API model cannot match The simultaneous launch across consumer, developer, and enterprise surfaces signals that Google views AI music generation as infrastructure, not a standalone feature.

Potential risks and opportunities

Risks

  • Rights holders tracking AI-generated music may challenge Google's artist-mimicry refusals as insufficient if style-based prompting produces recognizable outputs, potentially triggering litigation before end of 2026.
  • ProducerAI and other third-party integrators face downstream copyright liability exposure if SynthID watermarks are stripped and Lyria 3 Pro outputs are redistributed commercially without disclosure.
  • Stability AI's open-weight Stable Audio 3 could capture enterprise customers who need on-premise deployment for content clearance workflows, shrinking Lyria 3 Pro's addressable market in media and advertising.

Opportunities

  • Music licensing platforms (Musicbed, Artlist, Epidemic Sound) could integrate Lyria 3 Pro or Stable Audio 3 to offer instant custom-tracked content, compressing their production costs significantly.
  • Audio provenance and rights-management startups (Pex, Audible Magic) gain leverage as SynthID adoption raises enterprise demand for watermark detection and verification tooling.
  • Google Cloud partners building on Vertex AI have an early window to productize structured song-section prompting into workflow tools for ad agencies and video production houses before the interface becomes commoditized.

What we don't know yet

  • Whether SynthID watermarks survive lossy compression formats like MP3 at common streaming bitrates has not been independently verified.
  • Pricing and rate limits for Lyria 3 Pro on the Gemini API relative to Vertex AI have not been disclosed in available documentation.
  • How Lyria 3 Pro handles songs that describe a named artist's style without using their name -- a known workaround -- has not been addressed publicly.