Google DeepMind launches Lyria 3.5 in Flow Music: better vocals, lyrics, and creative control

AI-generated music has long had a recognizable flaw: flat melodies, robotic vocals, lyrics that feel like filler. Lyria 3.5, Google DeepMind's new music model that landed inside Flow Music on July 29, targets exactly those weak spots.

The declared advances touch four areas: musicality (richer, more natural melodic structures), lyrics (better adherence to the prompt and awareness of song structure — verse, chorus, bridge), vocals (more expressive, emotionally nuanced, better pronounced singing), and creative control, with easy adjustment of tempo and duration.

That last point is the most practical: you can request a track that lasts exactly as long as you need — a 60-second clip for a video, a 90-second background bed, a full 3-minute song — without manual cuts or looping. All of it is included for anyone using Flow Music, at no additional cost.

It is the third Lyria release in roughly five months (Lyria 3 in February, 3 Pro in late March), and the pace tells a story: while some music-AI startups fight record labels in court, Google keeps embedding generative music into its video-creation tools, where a made-to-measure soundtrack is a feature, not a standalone product.