The first time I put a song idea into Suno and got a finished-sounding track back in about thirty seconds, I sat there a bit stunned. It was genuinely impressive, and I think anyone who tells you it isn't hasn't really used it. But the more I listened back to what it had given me, the more I could hear that something wasn't quite right. It sounded almost like a proper record, but there was a smoothness to it, a kind of plastic sheen sitting over everything, and once you've noticed that you can't stop hearing it. It's the thing that makes people go "oh, that's AI" before they could even tell you why.
For a long while I assumed the answer was to keep working on the prompt, that somewhere out there was the right combination of words that would make it sound human. It isn't, and I wasted a fair bit of time finding that out. What actually fixes it is much simpler, and it's the one thing the AI genuinely can't do for you. You put yourself back on top of it.
The biggest single thing you can do is re-sing the lead vocal in your own voice, over the top of the AI track. A real human voice, recorded properly, is the thing that stops a track sounding generated more than anything else, because the vocal is the part our ears are most tuned in to. It doesn't have to be a perfect performance. It just has to actually be you, with your breath and your phrasing and all the little imperfections that a machine smooths away.
The next thing, if you want to take it further, is to replace one of the parts. You can pull an AI track apart into its separate pieces now — the vocals, the drums, the bass and so on — and swap one of them out for something you've played yourself. Re-record the guitar, or put a real bassline under it, so that the front of the track is genuinely your playing rather than a generated approximation of it.
And the last thing is to mix it yourself, rather than living with the flat, loud bounce the AI hands you. The balance it gives you is fine for a quick listen, but it's generic by design. Your own EQ, your own sense of where things should sit, the space you give the vocal — that's the difference between something that sounds like a demo and something that sounds like a record.
The way I think about it now is that the AI is a fast session band that'll play whatever you ask for in seconds, which is brilliant. But a band still needs someone out front, and that someone is you. Let it do the heavy lifting, then put yourself on top of it, and that "oh, that's AI" feeling goes away.
The full workflow — how to record the vocal so it actually sits, when pulling a track apart is worth the bother and when it'll bite you, and where AI earns its place versus where it can't — is in The Recording Manual.
If any of this is useful to you, come and join The Soundcheck — my newsletter, where I send out new guides and the odd story from the road.
See you out there.
— Aaron