AI Pulse by Inblix

Google's Lyria 3.5 can now surgically edit AI tracks — but still won't say what it trained on

The Decoder · Jul 29, 2026 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: Google's Lyria 3.5 can now surgically edit AI tracks — but still won't say what it trained on

Google dropped Lyria 3.5 on Wednesday, an update to its music generation model that brings a genuinely useful trick to the table: you can now edit individual sections of a track without trashing the whole thing. The feature, called Selective Section Painting, lives inside Google Flow Music and lets users isolate and tweak specific parts — a vocal line here, a drum fill there — or stretch a 30-second sketch into a three-minute arrangement. That’s a workflow upgrade that actually matters if you’re building songs iteratively instead of just rolling the dice on a prompt.

Sonically, Google says 3.5 delivers more natural melodies, tighter lyrics, and vocals that don’t sound like a robot with a head cold. Better pronunciation, more realistic timbre. Users get finer control over tempo and can set track lengths anywhere from half a minute to three minutes. It’s a clear shot at Suno and Udio, both of which have been iterating fast and grabbing mindshare among AI-curious musicians. The question is whether “sounds better” translates to “people will switch tools for it.” Musicians are sticky with their workflows.

Here’s where the needle skips. When Lyria 3 launched, Google vaguely gestured at training on materials it had rights to under YouTube’s terms and partner agreements — a sentence so carefully lawyered it could mean anything. Ask about 3.5’s training data and the company goes quiet. No response at press time. For a model that generates vocals mimicking human singers and instrumental performances, that silence is going to grate on artists who’ve already made their feelings about unlicensed training abundantly clear.

I’d argue Selective Section Painting is the real story here, not the fidelity bump. AI music tools have been a slot machine: pull the lever, get a song, hate the bridge, pull again. Being able to keep 80% of a generated track and surgically fix the rest makes the tool compositional rather than purely generative. That’s the difference between a novelty and something you might actually release. Whether Google can keep shipping features faster than the copyright questions catch up is the bet they seem willing to make.

💡 Key Takeaways

  1. Selective Section Painting lets users edit specific parts of AI-generated tracks instead of regenerating from scratch, shifting the tool from a slot machine to a compositional instrument.
  2. Google claims Lyria 3.5 produces more natural vocals, better lyrics, and clearer pronunciation, but provided no specifics on how these improvements were measured or benchmarked.
  3. The company declined to disclose Lyria 3.5's training data, repeating a pattern of opacity that will intensify legal and artistic scrutiny as the model's output becomes more convincing.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

Learn more

Glossary terms

← Back to all articles