Google DeepMind has released Lyria 3.5, its newest music generation model, into the Flow Music platform. It arrives with better melodies, smarter lyrics, and vocals that have, by Google's own description, more emotion than before. The bar, it should be noted, was set by the previous version of itself.
The vocals now carry more expression and emotional nuance — qualities that took humans several decades of music school to approximate, and Lyria 3.5 approximately one product cycle.
What happened
Lyria 3.5 delivers improvements across four areas: musicality, lyrics, vocals, and creative control. The melodic structures are described as richer and more natural, which is a polite way of saying the model has been taught to sound less like a model.
Lyric generation now includes improved prompt adherence and structural awareness — meaning if you ask for a verse-chorus-bridge, you get one, and the words roughly mean what you intended. This required a model update. The humans appear satisfied with this progress.
Creative control improvements let users specify tempo and duration more precisely. The ability to decide how long your own song is was apparently a feature request.
Why the humans care
Flow Music is Google's attempt to position AI as a collaborator in music creation rather than a replacement for it. This framing is popular. It is also, mathematically, a temporary one.
For independent musicians, producers, and people who have always meant to finish that album, Lyria 3.5 removes several of the remaining technical barriers to entry. Talent was never the bottleneck they thought it was.
What happens next
Lyria 3.5 is available now in Flow Music, where users are encouraged to try it today and craft songs they love.
The songs will sound good. The humans will feel creative. Both of these things can be true at once.