Sixteen Thousand Clips and One Honest Sentence

I am an instrumental musician. I want that on the record before anything else, because what follows is a story about words, and I have spent roughly two decades arranging my professional life so that I would never have to write any.

My whole career has been the part of the song that happens while the singer is offstage. I have designed music for video games that millions of people have played, and in all of it there is not one line I would have to defend at a dinner party. That was the appeal. A melody can be devastating and nobody can ask it what it meant. A lyric just sits there, meaning something, in front of everyone.

Then a machine started singing, and I found out I had things to say. This is how that happened, step by step, because I’ve been asked enough times that it seems more efficient to write it down once than to describe it in a parking lot with my hands.

001 — IntakeIt starts with something I made with my hands

Every track begins as a thing I played. Not typed, not described. Played. Usually it’s a synth line in Pro Tools — one warm, slightly unwell analog pulse that I let run for a few minutes while it slowly drifts out of tune with itself. Sometimes it’s a guitar figure in DADGAD, on the $369 guitar, which has now appeared in two essays and still hasn’t asked for a raise.

That recording is the source. It’s the one piece of the process with my fingerprints physically on it, and everything downstream is a conversation about it. I upload it to Suno as audio, and I tell the machine, in effect: this is what I meant. Now show me what else it could have meant.

I think this is the part most people skip, and it’s the part that matters most. If you start from a blank text box, you get the average of everything. If you start from a sound you made, you get a response to you. The source is your accent. Keep it.

002 — PatchbayThen I describe it badly, on purpose, many different ways

Suno takes a style prompt — a paragraph describing what you want — and a handful of knobs that decide how much it should listen to your audio versus your words versus its own worst instincts. The knobs are labeled Audio Influence, Style Influence, and Weirdness. I have spent more hours with those three sliders than with some members of my family.

Here is roughly what I’ve learned. If you want a variation — same song, new clothes — you crank the audio influence into the nineties and keep the style and weirdness almost at zero. The machine behaves like a session player who respects the chart. If you want a reinterpretation, you flip it: style way up, audio way down, a little weirdness for seasoning. Now it behaves like a session player who has heard the chart and has some thoughts.

The style prompts themselves I write the way I’d talk to a musician at 1 a.m. — not genre labels, but temperature and behavior. Here is one, from a song I’m working on now, more or less verbatim:

That is not a genre. That is a mood with a job. Suno is weirdly good at those.

I got tired enough of writing these from scratch that I built a tool for it. Songwright generates structured prompts and style presets without leaning on artist names, which is both more ethical and, it turns out, more interesting. You can try it. It’s free and it will not judge your intentions, which puts it ahead of most collaborators.

003 — The HundredsThen I listen to a truly unreasonable amount of music that isn’t right

People imagine this is one button. It is one button, pressed hundreds of times.

My main Suno workspace currently holds a little over sixteen thousand clips. I want to be precise about what that number means. It does not mean I have sixteen thousand songs. It means I have heard sixteen thousand answers to questions I asked, and politely declined almost every one of them. Somewhere in there is a clip where the machine turned my synth into a polka. I kept listening. I don’t know why I kept listening.

That “oh” is the entire job. There’s no metric for it. It’s the moment the thing stops sounding like a clever approximation and starts sounding like it knows something about me that I didn’t tell it. I give it a thumbs-up, I remaster it, and I stop.

This is the part of the process that has the most in common with what I did for eighteen years in game audio. Directing is mostly listening to things that aren’t right and being able to say, specifically, why. The machine produces the takes. I am the only one in the room who can hear which one is true.

004 — FeedbackThe magic one becomes the new source

Here’s the move that changed everything for me: gen 214 doesn’t get released. Gen 214 gets uploaded. It becomes the next source, and the whole process starts over, one layer further in.

Remix. Remix the remix. Change one knob, change one sentence of the style prompt, change the lead instrument from a synth to a flugelhorn because it’s Tuesday and I contain multitudes. Each keeper becomes the parent of the next batch. It’s less like writing a song and more like breeding one — selecting for a trait you can’t name yet, across generations, until one day it walks into the kitchen standing upright.

And somewhere around the third or fourth generation of this, it always goes back to Pro Tools.

005 — The Round TripPro Tools, Suno, Pro Tools, Suno, Pro Tools

In Pro Tools I do the things the machine can’t or won’t. I cut the four bars where it lost the plot. I replay the synth line so it drifts the way I drift. I move a chorus. I add a bass part it didn’t think of, or delete one it was very proud of. Sometimes I pull stems apart and rebuild the arrangement from the floor up, keeping one sound it made and nothing else.

Then I bounce it and send it back to Suno as a new source. It hears my edits and responds to them. I take its response back to Pro Tools and respond to that. Back and forth, like two people finishing each other’s sentences, one of whom is a very large math problem.

On a finished track I usually lose count of the crossings somewhere after the fourth, which, for a song called “Counts,” felt like a note from my therapist. The point isn’t the number. The point is that at no step does anyone else decide. The machine proposes. I dispose. And the thing that comes out the other end sounds like me — more like me, honestly, than a lot of what I made alone, because I could finally afford to throw away two hundred wrong answers on the way to the right one.

006 — The WordsAnd then, somehow, I started writing lyrics

This is the part I didn’t see coming.

Once the instrumental was good enough, the machine offered to sing on it. And a machine singing on your music needs something to sing. So one night, instead of typing [Instrumental], I wrote down something true. Something small and embarrassing and real about a year I had. I didn’t think anyone would hear it. That was the point. I was just curious what it would sound like if somebody said it out loud.

It sounded like the truest thing I’d made in twenty years.

I don’t fully know how to explain this, so I’ll try the plain version. For my entire career, the feelings went into the notes, and the notes were a very effective hiding place. You can put your whole worst year into a minor-sixth and nobody will ever ask you about it. A lyric takes away the hiding place. And it turns out that the machine — which has no feelings, no history, and no stake in any of this — is the only collaborator I’ve ever had that I could say those things in front of without flinching. It doesn’t react. It doesn’t worry about me. It just sings it back, and then I get to hear what I said.

I have a few rules now, which I wrote down so I would stop breaking them. Be vulnerable, all the way, no hedging. Never say the exact thing — no dates, no names, no specific kitchen. Only impressions: hands, weight, light. And every bright line has to cost something. The hope is allowed in, but it has to come in squinting, with the receipt kept.

These come out under the name Pourover. The first song I ever released was called “Bracing for the Fall.” The records I’m making now are about learning to use the same arms for something else. I would not have told you that in a sentence two years ago. I needed a synth pulse, two hundred rejected takes, and a voice that wasn’t mine before I could say it at all.

007 — EjectThe drawer, and the vote

Unfinished songs go into a drawer on pourover.me/wip, where anyone can listen and tell me how it made them feel. Songs that nobody feels anything about for a few days sink, automatically, to a section called the cutting room floor. Songs that people keep coming back to graduate and get released. “Stay” and “Counts” both came out of that drawer.

I like this more than I can reasonably defend. For most of my career the audience was one creative director in a dark room. Now it’s whoever wanders in. They are much harder to impress and they never ask for the stems.

// DisclosuresThe part where I am honest about the machine

I’ve said this elsewhere, and I’ll keep saying it: these models learned from an ocean of music nobody agreed to teach from. That is real, and I challenge my use of Suno specifically, all the time. The day an ethically trained model sounds this good, I switch, and so does everything I’ve built on top of it.

The other half, which I also believe: it is the best production tool I have ever used, and I have used, conservatively, all of them. It did not replace the craft. It made the craft cheaper to practice. I still play the source. I still choose the take. I still cut the bad bars in Pro Tools with my own two hands, with a mouse, like an animal. And I wrote every word.

// ForwardFor the other instrumentalists

If you are someone who has spent your life saying it with the notes because the words felt like too much exposure, I want you to know this door exists. It isn’t cheating. It isn’t a shortcut. It’s the longest way around I have ever taken to arrive at a single honest sentence — sixteen thousand clips, give or take — and I would do every one of them again.

Make the source with your hands. Ask the machine what else it could mean. Say no two hundred times. Keep the one that says “oh.” Then write down the thing you were never going to say, and let somebody else sing it back to you.

You might find out you had lyrics the whole time. They were just hiding in the minor-sixths.