Lyria 3.5 Lets You Rewrite One Section of a Song

Most AI music tools have shared a single frustrating property: you get a whole track or nothing. Dislike eight seconds of the chorus and your only option is to regenerate everything and hope the parts you liked survive. Google's Lyria 3.5, which began rolling out in Flow Music on July 29, goes after exactly that problem.

Selective Section Painting is the real feature

The headline capability is called Selective Section Painting. It lets you edit a specific portion of a track without starting over, and it works in the other direction too: hand it a short melody and it can extend that into a full song.

This matters more than the model quality improvements shipping alongside it. Generative audio has been stuck at the slot-machine stage, where iteration means rerolling and comparing. Editing a region while holding the rest fixed is what turns a generator into a tool someone can actually work with.

The rest of the release is incremental and reads like a normal version bump. Google cites richer melodic structure, better lyric quality with improved prompt adherence, more realistic and emotionally nuanced vocals, better pronunciation, and direct control over tempo and duration. Output length runs from 30 seconds to three minutes.

Does anyone actually need generated music?

Worth being honest about the ceiling here. Nobody is replacing a composer whose work carries a project, and the three-minute cap tells you where Google thinks this lives.

The real demand is duller and larger: background audio for product videos, social clips, in-store playlists, app sounds, podcast beds. Work that currently means either paying for a stock music licence or paying someone a modest fee to produce something forgettable. That is a genuine market, and section-level editing is what makes it addressable, because that kind of work is almost always "nearly right, fix the middle eight."

Tempo and duration control point at the same use case. If you are scoring a 22-second product clip, matching the cut is the whole job.

Before you put generated music in anything commercial

Two cautions, neither of which Google's announcement resolves for you.

The first is licensing. Whether generated audio can be used commercially, and under what attribution or platform terms, depends on the specific product tier you are on. Read those terms before the track ships in a campaign, not after.

The second is platform detection. Social platforms and streaming services are actively building AI-content identification and their policies are changing quickly. Audio produced today under one set of rules may be labelled differently a year from now. For a short-lived ad this is irrelevant. For evergreen brand assets it is worth a moment's thought.

The pragmatic position: this is now good enough for functional background audio, and section editing removes the workflow objection that kept it out of production. It is not a reason to cancel a relationship with anyone whose music people actually notice.

Sources: Google Blog, The Decoder, Google DeepMind, Android Authority