Google's Lyria 3.5 adds section editing to Flow Music
Google launched Lyria 3.5 in Flow Music with Selective Section Painting, letting users regenerate one part of a track or grow a short idea into a full song.
Google launched Lyria 3.5, the latest version of its music generation model, inside Google Flow Music on Wednesday, with the headline change being editability rather than raw fidelity.
The new Selective Section Painting feature lets users regenerate a chosen portion of a track — a verse, a chorus, a bridge — without discarding the rest, or take a short melodic sketch and extend it into a complete arrangement. Generations run from roughly 30 seconds to three minutes, and the release adds finer control over tempo and over individual stems such as vocals, drums and bass.
Google also lists improvements across the model itself: richer and more natural melodic structure, lyrics that follow prompts more closely and show better song-structure awareness, and vocals that are more emotionally expressive with clearer pronunciation. Output duration control is expanded as well.
The company has not disclosed what Lyria 3.5 was trained on. When it shipped Lyria 3, Google said the model was trained on material that YouTube and Google had the right to use under their terms of service, partner agreements and applicable law, without naming datasets or rights-holders; it did not answer questions about the newer model's training corpus.
Section-level regeneration is the feature commercial music tools have been converging on, because it addresses the structural weakness of generative audio: a single-shot generation is a slot machine, and a producer needs to fix the second chorus without losing the first. Suno and Udio have both moved in this direction, and stem-level control is the prerequisite for AI-generated audio to enter conventional digital audio workstation workflows rather than sitting beside them.
Why it matters: Music has been the slowest of the major generative modalities to become professionally usable, and the reason has been control, not sound quality. Putting section-level editing and stem handles into a Google-distributed product moves AI music from novelty output toward material an editor can iterate on — while leaving the unresolved question of training provenance, which is the axis on which the industry's licensing fights will be decided, untouched.