Crate Digging for Prompts

Crate Digging for Prompts

Collecting prompts is crate digging, and the DJs figured this out fifty years ago

I was fourteen when I started digging, and there were exactly two record stores downtown. That was the whole world. Two rooms, and a number of crates between them I could probably still count, and I went through all of it. Not once. Repeatedly. The way you re-read a book you already know the ending of, because the fifth pass is when you start noticing things.

I wasn't shopping for records. I was hunting sounds.

Scratch DJs dig differently than producers do. A producer at least wants a bar or two, something with a shape to it. I wanted single hits. A horn stab. One vocal syllable clipped so hard it stops being a word. A zap, a siren, a laugh, an ahh I could pull back and forth under my hand until it stopped being a sample and turned into an instrument. Then, in the same bins on the same afternoon, I'd be listening for something completely different. Four bars of drums, a bassline worth looping, something to hand to whoever needed a beat that month. Rappers. Bands. Whatever was in front of me.

Two stores. Hours per visit. Most trips, nothing. Some trips, one sound that paid for the whole year.

Then I'd file it and wait until I needed it.

That last part is the part nobody talks about, and it's the part that matters here. Because if you have ever screenshotted someone else's prompt off a feed at 1 a.m. and dropped it in a folder for a project that doesn't exist yet, you have done the exact thing I was doing in those two rooms. You just haven't been told it has a history.

You're not looking for the song

The thing that separates a digger from a shopper is what they're listening for. A shopper wants the record. A digger wants four bars of it: the moment the band drops out and the drummer is alone, the horn stab, the half-second of a vocal ad-lib that can be chopped into something else entirely. The rest of the record is packaging.

The Amen break is six seconds long. It's a drum solo buried in the B-side of a 1969 gospel-soul single by The Winstons, a record that went nowhere. Those six seconds became the structural foundation of jungle, drum and bass, and a decent chunk of hip-hop. Nobody was digging for "Amen, Brother." They were digging for the part where everything else stops.

Prompt collecting works the same way, which is why "I saved the whole prompt" is a beginner's move. You are not collecting the prompt. You are collecting the one clause inside it that's doing all the labor. Cinestill 800T. chiaroscuro. shot on a 35mm lens, f/1.4. bas-relief. risograph, two-color misregistration. Everything around it is the song. That fragment is the break.

Once you start reading for the break instead of the prompt, the whole practice changes. You stop hoarding paragraphs and start collecting parts.

The label comes off

Here's the detail I can't stop thinking about. In the early Bronx, DJs peeled the labels off their records or covered them in tape so that rivals leaning over the booth couldn't read what was spinning. Kool Herc did it. Others soaked their records in the tub to get the paper off clean. The knowledge of what the record was had more value than the record.

That is a perfect description of the modern prompt economy and I don't think it's even a metaphor. It's the same behavior with the same motive.

People crop the metadata out of their exports. They post the image and not the recipe. They sell prompt packs on marketplaces, which is a business model that only makes sense if you accept that a paragraph of English can be a trade secret. There are Discord servers where the actual currency is the phrase you haven't heard yet.

And the same thing will happen that always happens: the break gets found, the break gets used, the break gets used by everyone, and then the break is dead. For two years in the style of Greg Rutkowski was a skeleton key that opened a specific kind of luminous fantasy rendering. Then it was in every prompt on the internet. Then it read as a tell, the visual equivalent of hearing the Amen break in a car commercial. Saturation kills a sample. It kills a modifier the same way, only faster, because there is no pressing plant slowing anyone down.

Everyone can buy the same records

The oldest insult thrown at DJs is that they didn't make anything. They just picked stuff other people made and played it in a room.

You have heard this insult recently, in a different outfit.

The rebuttal has not changed and it does not need to. Two people can own identical crates and one of them empties the floor. Selection is a skill. Sequencing is a skill. Knowing that this thing goes next to that thing, that the energy has to drop here so it can lift there, that the crowd in front of you tonight is not the crowd from Thursday. None of that is contained in the records.

And in practice nobody plays the sample straight anyway. You chop it, pitch it, filter it, loop the tail, layer it under something that has no business being there. The flip is the craft. The dig only supplies raw material.

Same with a prompt. The people doing interesting work aren't running found prompts unchanged. They're grafting the lighting language out of one onto the composition logic of another, keeping a texture term that has nothing to do with the subject, deliberately breaking a phrase to see what the model does with the wreckage. The collection is the sample library. What you do to it is the record.

The blindfold

There's a video series that makes this argument better than I can, because it isn't an argument. It's an experiment.

Mass Appeal's Rhythm Roulette has been running since 2013 and is well past a hundred episodes. The format never changes: a producer is blindfolded, walked through a record store, and pulls three records at random off the shelves. Then they go back to the studio and have to build a beat out of only those three.

It took me an embarrassingly long time to notice what that blindfold is actually doing.

It deletes the dig. Everything I spent my teenage years developing gets switched off. Knowing the bins, knowing the labels, knowing which shelf in which store rewarded a fourth look. Whatever the producer walks out with is random in the strict sense. In one early episode, 88-Keys ends up with a Steve Martin standup comedy album, which contains no music of any kind, and a hardcore noise metal record. That's the hand. Make a beat.

And they do. Consistently. The show would be unwatchable otherwise. If the collection were the skill, the premise would have collapsed in episode one, because a producer separated from their crate would be a producer with nothing. Instead you get the opposite. Strip out the acquisition and every ounce of the craft relocates to the transformation. You watch someone find the one usable half-second in a comedy record, a breath or a room tone or a consonant, then pitch it into the basement and loop it until it's a bassline with no memory of where it came from.

Now watch the El-P episode with that in mind. El-P has one of the most identifiable signatures in rap production: dense, mechanical, claustrophobic, closer to a malfunctioning spacecraft than a dusty soul loop. Hand him three records he has never seen, selected by chance, and the beat still comes out sounding unmistakably like El-P. Different episode, different producer, equally random material, completely different result, and each one still legible as its maker.

That is the whole rebuttal, and it's empirical rather than rhetorical. Randomize the inputs and the authorship survives. Which means the authorship was never sitting in the inputs to begin with.

So run it on yourself. Your prompt library has a problem you almost certainly can't see from inside it. It's a record of your past taste, so it will keep quietly steering you back toward work that resembles your old work. The feeds homogenize everyone globally. Your own crate homogenizes you, specifically, and it's the more dangerous of the two because it feels like having a style.

Prompt roulette, then. Pull three fragments out of your library at random: one light, one surface, one compositional rule. No regard for whether they have any business together. Force them into a single image. Don't rescue it when it goes sideways; build with whatever comes out. Better version: take the blindfold literally and pull your three terms from a field you know nothing about.

The results won't necessarily be good. That's not the point. The point is that constraint is the only dependable way out of your own preferences, and randomness is the cheapest constraint on the shelf.

The crate is the instrument

Here's where the metaphor stops being flattering and starts being useful.

DJs do not organize records by artist. They organize by BPM, by key, by energy, by when in the night this works. The filing system is built around the moment of use, because a record you can't find in ninety seconds under club lighting does not functionally exist. The crate isn't storage. It's an instrument you play under pressure.

Almost everyone organizing prompts is doing it wrong in exactly the way a novice organizes records: by subject. A folder called portraits. A folder called landscapes. A folder called logos. This tells you what the prompt was about, which is the least useful fact about it, because you will never again need that specific thing.

Organize by function instead. What does this fragment do?

  • Light: how it falls, how hard, what color, what time of day
  • Surface: grain, gloss, weave, wear, the physics of the material
  • Camera: lens, distance, height, movement, what the eye is doing
  • Composition: where things sit, negative space, symmetry, crop
  • Register: the mood, the era, the emotional temperature
  • Failure: the terms that reliably break things in a way you like

Now your collection is cross-cuttable. You can pull a lighting move you found in a food photo and put it on an architectural render, because you filed it as light rather than as lunch. That is the whole game. That is a DJ pulling a two-second bassoon from a library music LP and putting it under a drum pattern.

Tag the source too. Not for credit exactly. For return. Diggers remember which shop, which bin, which seller. If a feed or a channel gave you three good things it will give you a fourth.

Where the record stops spinning

I want to be honest about where this breaks, because an essay that only says X is like Y for two thousand words is a party trick.

Your crate can spoil. A record from 1969 sounds the same today as it did the day it was pressed. A prompt does not. Models update and the same text produces a different image. Sometimes a better one. Often just a stranger one. Anyone who built a careful library under one model version and watched it go sideways after an upgrade has experienced something no vinyl collector ever has. Your crate is renting its meaning from someone else's infrastructure. Build accordingly: save the outputs alongside the inputs, so you at least keep the evidence of what a phrase used to do.

The dig has been flattened. Crate digging was geographically constrained, and that constraint was the entire engine of regional style. Two stores downtown felt like a poverty at the time. I've come to think it was the best thing that could have happened to me: a small enough world that I had to go deep instead of wide, and a different small world than the kid doing the same thing three states over. What you could find in Detroit was not what you could find in Bristol, and that's why Detroit and Bristol sound different. The prompt dig happens inside recommendation systems that show broadly similar things to everyone. Everyone's basement is the same basement now. If you want a distinctive collection you have to go somewhere the feed isn't. Old photography manuals, printing trade vocabulary, materials science, film production glossaries, a language you don't speak. The obscure terminology of an unrelated field is the last real regional pressing.

And you can't name the source. A sample points at a record. You can find the record, buy it, hear the whole thing in context. A prompt points at an aggregate: a statistical smear of millions of images, most of which nobody consented to hand over. The digger's ethic was always partly about reverence. You knew whose drums those were, you knew you owed something. That thread doesn't attach to anything here, and pretending otherwise is the weak point of the whole analogy.

Dig toward a set

The failure mode of every collector is the same in both worlds: the crate becomes the hobby. You accumulate, you organize, you re-organize, you feel productive, and you never play out.

Records are for a room. Prompts are for a thing you're actually making. Both collections are worthless as collections and only become anything when someone stands in front of a crowd and takes the risk of putting the needle down.

Go dig. Then go play.