Google Launches Lyria 3.5 Music Model With Section-Level Editing, No Full Regeneration Required

Google released Lyria 3.5, an update to its music generation model that adds the ability to edit specific sections of a track without regenerating the entire piece.

The headline feature, called "Selective Section Painting," lets users modify individual parts of a song—or expand a short melody into a full track—while leaving the rest untouched. Users can also fine-tune the tempo and duration of individual elements including vocals, drums, and bass. Previous-generation tools typically required a full regeneration to change any part of a track, making iterative editing slow and unpredictable.

According to Google, Lyria 3.5 produces more natural-sounding melodies, improved lyrics, and more realistic vocals with clearer pronunciation compared to its predecessor. The company also says users now have more precise control over tempo and track length. Generated tracks can run from 30 seconds up to 3 minutes.

Lyria 3.5 is available now through Google Flow Music, Google's consumer-facing music generation product.

Training data remains undisclosed

When Google launched Lyria 3, the company said the model had been trained on material that YouTube and Google had rights to use under their terms of service, partner agreements, and applicable law—without specifying which artists, labels, or catalogs were included. Asked about the training data behind Lyria 3.5, Google did not immediately respond, leaving the same ambiguity in place for the new version.

No benchmark scores, parameter counts, or pricing details were disclosed alongside the release. Google has not published independent evaluations comparing Lyria 3.5's output quality against competing systems from Suno or ElevenLabs.

What this means

Section-level editing addresses a real workflow gap in AI music generation: until now, most tools treated a generated track as a single indivisible output, so fixing one bad vocal line or drum fill meant re-rolling the whole song and hoping for a better result. Letting users isolate and adjust vocals, drums, bass, and tempo independently moves Lyria closer to how producers actually work in a DAW, and it's a meaningful step toward AI music tools being used for iteration rather than one-shot generation.

The unresolved training-data question matters more as these tools get more capable. Google's refusal to detail what Lyria 3.5 was trained on—repeating the same vague YouTube/partner-agreement language used for Lyria 3—keeps the model in a gray zone that has already drawn scrutiny from artists and labels. As AI-generated music becomes more editable and production-ready, pressure on Google to clarify licensing terms, and on regulators to weigh in, will likely increase rather than fade.