I set out to make a classic soul piece. Sweet Magic had other ideas, but I still had to decide which ones were worth following.
The song developed through a conversation between prompting, listening and rewriting. Each result suggested something: a change to a lyric, a different shape for a section, another direction for the sound. As the words and structure came together, I began to hear possibilities beyond the soul piece I had originally imagined.
That sounds like freedom. It is also a very efficient way to spend an entire evening making decisions you thought you had already made.
“Pretty good” is where the difficult work starts
When the lyrics and structure are nearly right, a song can work surprisingly well in several styles. You can get a convincing result in one direction, try another, and get something convincing there too.
Now what?
Having more good options does not tell you which one belongs to the song. That is where taste becomes a responsibility. You need a goal, and you need to push towards it. The goal can evolve as you discover something better, but every attractive result cannot become the new destination.
Otherwise, you are collecting versions rather than finishing a record.
The producer is still in the room
Listen critically. Does the phrasing serve the lyric? Does the structure hold together? Is that strange little sound an interesting detail, or a glitch you have heard so many times that you have started defending it?
A glitch does not become artistic intent just because you are tired.
Change the prompt. Try again. Listen again. Decide whether the adjustment actually helped. A new version is not automatically a better version, and something impressive is not necessarily something you should keep.
I sometimes imagine doing this with a human band.
“Lovely take. Could we do it again, with one tiny change?”
By the hundredth request, the drummer would be packing up, the singer would have blocked my number, and the bass player would be asking whether this rehearsal was paid.
AI has yet to storm out of my studio. That does make it easier to ask for another attempt. It does not tell me whether the attempt is worth keeping.
Yes, I pressed a button. Then I had to make a decision.
There is a familiar assumption that making music with AI means typing a sentence, accepting whatever comes back, and calling yourself an artist.
You can work that way. You can also accept the first draft of a lyric. I would not recommend either as a universal quality standard.
With Sweet Magic, the work was in developing the words and structure, listening to what each result revealed, and deciding what to change. The tool generated possibilities. I remained responsible for choosing among them.
That does not mean effort alone makes a song good. Nobody owes me a second listen because I rejected a hundred versions. If the final recording does nothing for you, the number of attempts is hardly a defence.
But dismissing it before listening because AI was involved is also a remarkably quick creative decision. Certainly quicker than mine.
Eventually, you have to call it finished
The same freedom that helps you develop a song can keep you trapped inside it. There is always another prompt, another variation, another small adjustment that might make everything click.
The challenge is recognising when you are still improving the work, and when you are simply postponing the moment of letting it go.
Finishing means making a decision you can stand behind. This is the version. This is what I want the song to say.
Then you release it. Other people get to hear it, enjoy it, dislike it, or find something in it you never intended.
And you get to move on to the next project.
Probably a classic soul piece.
We will see.
Listen to Sweet Magic
Here is the version I chose to release. Listen, then decide for yourself.