Dheeraj Kumar

One more thing I've been using that doesn't get talked about enough — AudioShake for stem separation when the track is commercially released and you need clean splits for sync or remix licensing.

It's different from ACE-Step or LALAL in that it's built for professional licensing workflows, not just DIY splitting. If you're working on a remix that needs to go through label approval or sync clearance, the output quality and the metadata it provides actually matters.

For visual stuff — if you're doing music visualizers for YouTube or social, Luma Dream Machine + your own audio reactivity layer in TouchDesigner is genuinely a strong pipeline right now. Not fully automated, but the quality gap over plug-and-play visualizer apps is massive.

Anyway — lots of good tools in the ecosystem right now. Most of them are genuinely useful if you know what problem you're actually trying to solve. Happy to go deeper on any of these if useful.

Suno vs Udio — real talk since everyone keeps asking.

Suno is the easier starting point. You put in a rough idea and it spits something out fast. Great for quick demo sketches when you want to hear a vibe before you commit to building it properly. The interface doesn't get in your way.

Udio genuinely has better vocal realism though. If the track you're generating has singing in it, Udio tends to sound less synthetic and more like an actual performance. The trade-off is it takes a bit more coaxing to get consistent results.

Honest take: neither of these replaces what you actually do as a producer. But using them as a sketching tool? Like, "does this chord progression and vibe even work before I spend 3 hours on it?" — that's where they're genuinely useful.

Play with both. They're cheap enough to try. What's your use case been for AI generation tools?

For producers who also code — whether that's plugins, Max for Live patches, or web projects — Claude Code has been my go-to this year and I think it's the strongest AI coding tool of 2026 so far.

What separates it from GitHub Copilot or regular autocomplete is that it actually finishes things. You describe a feature, it builds it. Not just the next line — the whole thing. It handles file structure, edge cases, and it'll tell you when something is a bad idea rather than just going along with it.

Best setup I've found: pair it with Cursor IDE. Cursor handles the file navigation and inline edits, Claude Code handles the heavier reasoning and multi-file tasks. They actually complement each other really well.

If you've been putting off learning to code because it felt too slow, honestly give it a shot with this combo. It changes the pace.

If you do remix or DJ edit work and you're still paying for stem separation — check out ACE-Step-1.5 on GitHub before you renew anything.

It's open-source, runs fully local on your machine, and does a decent job isolating vocals, drums, and bass. No cloud uploads, no subscription, no data going anywhere you don't control.

Not gonna pretend it's perfect — LALAL.AI and Moises still edge it out in tricky mixes — but for everyday stuff like pulling a vocal for a quick edit or grabbing the kick layer from a track? Honestly solid. And free is hard to argue with.

Anyone else been using open-source tools for this kind of thing? Curious what setups people are running.

Been doing some deep dives into AI mastering lately and honestly iZotope Ozone 12 vs LANDR Pro is a pretty interesting comparison worth talking about.

Ozone 12 still wins for me when it comes to control — especially the Stabilizer module. It basically tames frequency build-up across the mix in a way that feels surgical without sounding over-processed. Really useful if you're working on dense electronic stuff.

LANDR Pro is more "fire and forget" and sometimes it totally nails it, but it can feel like a black box. Futureproof Music School did a solid breakdown of both if you wanna go deeper — worth checking out before you commit to a subscription.

For me: Ozone 12 for anything client-facing, LANDR for quick references and rough demos. What are you lot using?