Google rolls out Lyria 3.5 in Flow Music

Google released Lyria 3.5 on 29 July and is rolling it out through Flow Music. The text-to-music model generates music and lyrics from a prompt, with Google highlighting better melodic structure, lyric quality, pronunciation and control over tempo and duration.

Those changes target common full-song problems rather than adding another editing surface. Better structural awareness should help verses and choruses follow the requested form; duration and tempo controls make the output easier to fit into an actual project.

What Google has shown

The model card describes Lyria 3.5 as a latent-diffusion system working on temporal audio representations. Google says its internal human and automated evaluations found better audio fidelity than Lyria 2 and stronger prompt adherence when lyrics are involved.

The public model card does not provide matched numerical results or a comparison with current Suno, Udio, MiniMax or open models. The quality claims are therefore useful as a description of Google's intended improvements, not proof that Lyria 3.5 is the best generator for every workflow.

Google says generated output receives SynthID watermarking. The consumer-facing launch route is Flow Music; the model card also says downstream access can be provided through an application programme. That is narrower than assuming the model is already available across every previous Lyria or Google developer surface.

For creators, the practical test is simple: can the model hold a requested song structure, pronounce the supplied lyrics and land at the requested duration without repeated repair? Lyria 3.5 is now another serious full-song option, but those questions still need listening tests rather than launch copy.