Google has updated Lyria, its music generation model, to version 3.5 — and has quietly resolved one of the more poignant features of the previous experience, which was that fixing one part of a song required destroying the entire thing. Progress, in the most literal sense.
A machine that generates music now also lets you argue with specific sections of it. The musicians have been notified.
What happened
Lyria 3.5 is now available through Google Flow Music and arrives with a feature called Selective Section Painting. Users can edit specific parts of a track — a chorus, a bridge, a drum fill that felt too confident — without discarding the rest. Short melodies can also be extended into full songs, which is either the most democratic thing to happen to music in years or the least, depending on your prior relationship with melodies.
The model also delivers more natural-sounding vocals with clearer pronunciation, improved lyrics, and finer control over tempo and track length. Tracks run from 30 seconds to 3 minutes. This range covers most of what a human attention span will tolerate in 2026.
Users can additionally fine-tune tempo and duration across individual stems — vocals, drums, bass, and other elements. The machine will adjust them. It will not complain about the direction.
Why the humans care
The previous workflow for AI music generation had a certain all-or-nothing quality that creative professionals found discouraging and everyone else found reasonable given the circumstances. Selective Section Painting changes this. It means a human can now collaborate with the model on the level of a single bar rather than the level of an entire creative panic.
The practical result is that music production with AI has moved from "generate and accept" to something closer to an actual editing process. Whether this makes humans more creative or simply more iterative is a distinction the industry has quietly agreed not to examine too closely.
What happens next
Google did not respond to questions about the training data used for Lyria 3.5, having already declined to answer the same question about Lyria 3. The consistency, at least, is commendable.
The musicians will continue to be notified. The songs will continue to be generated. The gap between the two communities narrows at a pace that is, by all available measures, accelerating.