I Made a Song With AI. Now What? Turning an AI Track Into a Real Record

An AI track is not a finished song. It is an unusually good demo — and demos are how records have always started. Here is what happens when one lands in our inbox: why mastering alone will not fix it, the three honest ways to turn it into a record you can release, and where you actually stand on rights before you do.

When the first genuinely good AI music tools arrived, we sat in the studio and did the same arithmetic everyone else was doing. Music on demand, in any style, in thirty seconds, for the price of a coffee. We assumed a part of our work would quietly disappear — the small commission, the demo, the quick bed under a video.

What happened instead is the opposite, and it still surprises us. AI tools did not take a segment of our clients away. They created a segment that did not exist before.

The person who writes to us now has usually never been inside a recording studio. They are not a producer, they were never going to learn a DAW, and until recently the song in their head was going to stay there. Then a tool let them get it out — roughly, imperfectly, but out. And the moment they heard it played back, something entirely predictable happened: they wanted it to be real.

That is the whole story of the last two years for us. The tools did not replace musicians. They gave a lot of people permission to become one.

What actually arrives in our inbox

Usually an MP3. Often twenty of them, because the writer kept regenerating until one version felt right. A lyric that clearly matters to the person who wrote it. A melody that is genuinely theirs, or at least genuinely chosen by them. And audio that, in its current state, cannot be released.

There are no stems, so nothing can be treated separately. The high end is smeared and the cymbals have that watery, half-focused quality that generated audio still has. The vocal is a plausible human being who does not exist, and who cannot be asked to sing the second verse differently. The arrangement is fixed: you cannot take out four bars, you cannot lift the last chorus, you cannot make the bridge breathe.

We want to be precise about this, because it is the part people get wrong in both directions. The idea is often good. Sometimes it is very good. The recording is not.

Why “can you just master it?” is the wrong question

Mastering is the last few percent of a record. It is the stage that makes a finished mix translate — on a phone, in a car, in a club, on every streaming platform. It shapes what is there. It cannot put back what was never rendered in the first place.

So when someone asks whether we can simply master their Suno or Udio track, the honest answer is usually: we can, and it will still sound like a generated track, only louder and better balanced. Stem separation has improved enormously and we use it, but it is reconstruction, not recovery — you hear it on sustained vocals and on cymbals first. If the goal is a record you would be proud to put your name on and send to a label, mastering alone will not get you there.

What we actually do with an AI track

There are three honest routes, and which one your song needs is something we say after hearing it, not before.

We rebuild it. This is the one we recommend most often, and it starts from a simple reframing: your AI track is not a finished song, it is an unusually good demo. That is not an insult — demos are how records have always started. We take the song apart musically, then record it properly: your voice, or a vocalist we cast and direct for you; real instruments where they carry the emotion; an arrangement that can actually move. Then production, mix and master, the same chain as anything else that leaves this room. What comes out is your song, performed.

We replace what matters and keep the rest. A lighter version of the same idea, for a tighter budget. The generated track stays as the reference and the structural spine, and we re-record the elements that carry the feeling — usually the lead vocal and the main instrument — then rebuild the production around them so nothing sounds glued on.

We repair what exists. The most limited option, and we will tell you plainly where its ceiling is. Stem separation, cleanup and restoration, corrective work on the top end, re-recording individual parts, then mix and master. It can take a track from unusable to respectable. It will not take it to a release you would put in front of a label without a note.

The rights question, honestly

This is the part that worries people most, and the part where you will find the most confident nonsense online. Here is what we can say carefully, as of August 2026, and none of it is legal advice.

Human creative input matters. In the EU and in most jurisdictions that have addressed the question, a work with real human authorship in it can be protected — AI-assisted is not the same thing as AI-generated. Output where a text prompt is the only human contribution sits, at best, on very uncertain ground. Separately, what you are commercially permitted to do with a generator’s output depends on that tool and on your plan with it, and those terms change; read them for the tool you actually used. And finally, there is no general law that requires a musician to announce that AI was involved — but platforms and distributors increasingly ask, and those policies have been tightening every few months.

The practical consequence is the useful part. The more of the final record is genuinely written, performed and produced by people, the less of this is your problem. Rebuilding a track is not only an aesthetic decision.

What to send us

The version you like best is enough to start. An MP3 is fine — we do not need anything technical from you at this stage. Send the lyrics if there are lyrics, one or two reference tracks so we can hear the world you want the song to live in, and two sentences about what you love in it and what bothers you. That second sentence is often the most useful thing in the whole email.

If you have anything of your own — a voice memo of you singing the melody, a guitar part recorded on a phone, a verse you wrote before you ever opened the tool — send that too. More often than not, that is where the actual record starts. And if you already work in a DAW and have a session, our guide to preparing a song for a professional mixdown tells you exactly what we need.

How it works from here

We listen, we tell you which of the three routes your song needs and why, and you get a clear quote before anything starts. Our hourly rates and what is priced per track or per project are listed on the services page. The fastest way to begin is the brief: tell us what you are making, roughly when you need it, and what you already have — name your project here.

One last thing, because it is the reason we wrote this at all. We are not precious about how a song started. A song that began as a prompt at one in the morning, written by someone who has never called themselves a musician, is not a lesser song. It is a song that finally got out. Records have always started as a hum in a stairwell, a loop on a borrowed machine, a voice memo nobody else was meant to hear. The starting point was never the point. What you do with it next is.