Google turns three-minute songs from thin air with lyria 3 pro

Google just handed musicians a console that spins full tracks from a handful of words. Lyria 3 Pro, quietly released this week inside Vertex ai and the Gemini app, stretches the earlier 30-second snippets into three-minute songs while letting users choreograph every intro, verse, chorus and bridge as if they were ghost-producing for themselves.

From prompt to stereo in one click

The jump is less about length and more about resolution. Where the base Lyria 3 filled gaps with algorithmic guesswork, Pro exposes the dials: tempo maps, vocal registers, stylistic pivot points, even bar-by-bar mood shifts. Type “1978 Tokyo disco, 110 BPM, female falsetto pre-chorus” and the model returns a fully arranged cut, stems included, ready for a mixing desk.

API calls cost the same, but the compute bill rises with complexity. A three-minute opus burns roughly 18× the tokens of a 30-second loop, pushing pricing into the realm of indie-label budgets rather than bedroom producers. Google banks on the bet that serious creators will pay for speed and control instead of looping short samples in a DAW.

Watermark buried in the wave

Every file sneaks out the door stamped with SynthID, an inaudible hash woven into the waveform. Labels and streaming bots can flag the fingerprint automatically, a move meant to keep ghost-written ai tracks from flooding Spotify before the industry agrees on royalty rules.

Early testers inside Google’s closed beta pushed the engine into stranger territory: generative score cues that evolve scene-to-scene for short films, multilingual jingles localized on the fly, even adaptive workout playlists that speed up when your smartwatch senses lagging cadence. Expect the first commercial wave this summer as marketing agencies swap stock-music subscriptions for prompt budgets.

The timing is strategic. Universal and Sony are negotiating new rate cards for machine-made music, and Google wants leverage: a fluent composer that can deliver broadcast-length material before rivals such as Stability’s Stable Audio or Adobe’s Project Music GenAI Control exit their own labs. If the majors flinch, Google already owns the pipeline.

For artists, the equation is brutal: automate arrangement or risk being undercut by someone who will. The three-minute single, once a month-long studio sprint, now compresses into the span of an espresso break. Labels won’t ask if you use ai; they’ll ask why you didn’t.