The Work Between the Renders
There's a conversation I have with other AI artists that goes the same way every time. Someone shows me a piece they're proud of, I ask about it, and they tell me the prompt. Then they apologize.
"I know it's just MidJourney," they say. "I just typed some words."
I want to talk about why that instinct is wrong, and what I actually mean when I say this is craft.
The prompt is the last ten percent
When I sit down to make a portrait, the prompt is maybe ten minutes of work. What happens before and after is hours.
Before: I'm thinking about the feeling I want to hold. A color I saw on a building yesterday. The way light broke across a specific face in a specific movie. A posture. A silhouette. The prompt is me trying to translate that feeling into words a model can parse.
After: I'm selecting. On a piece I'd actually publish, I've usually generated 40 to 200 images. Most are wrong. Some are close. A few are strange in ways I didn't expect. I pull the ones that move me. Then I start working on those — composite passes, color grading, masking out something that doesn't belong, sometimes repainting sections entirely. I'll often run a final image back through image-to-image at low strength to unify it.
The prompt is the door. It's not the room.
Your first hundred images aren't your style
This is the part I wish someone had told me earlier.
When you start, everything you make looks like MidJourney — or Flux, or SD, whichever model you picked up first. That's not a failure. That's just the baseline aesthetic of the tool you're using. Your "style" is what emerges once you've made so many images that you've developed strong opinions about what you keep.
Selection is a skill. It's the photography part of AI art.
A street photographer shoots 500 frames in an afternoon and publishes three. Nobody accuses them of cheating because the camera did most of the work. The craft is in where they pointed it, when they pressed the button, and what they kept. AI art is the same discipline, applied to a different kind of camera.
If you're frustrated that your work still looks generic after your first hundred images: you're fine. You're in the prompt-and-pray phase. The style comes after you've selected enough that your taste calcifies.
A small practical thing
If you're going to take one concrete thing from this, take this: don't publish the first good image you make on a given idea.
When I'm chasing something, the first "wow, this is it" image is almost never the one I publish. It's the one that showed me what I was actually after. I save it, I re-prompt with more specificity in the direction it pointed me, and I keep going. The piece I publish is usually the third or fourth "wow" moment — after I've learned what the subject wants to be.
You'll know you're getting somewhere when your selection criteria sharpens. When you can look at ten near-identical outputs and say "this one" without hesitating. That's your taste. Trust it.
On the discomfort of iteration
I know people who won't publish an AI piece because they're waiting for the definitive version. There isn't one. There never is.
A painter doesn't stop painting because the last one wasn't perfect. A writer doesn't stop writing because the last sentence was clumsy. You make the piece, you let it go, you make the next one. The work isn't the individual artifact — it's the practice.
This is the permission I wish someone had handed me earlier: you are allowed to publish work you will later grow out of. In fact, that's the only way forward. The artist you're going to be in a year is only reachable by shipping the work of the artist you are now.
Where I keep going wrong
To close with something honest: I still overwork pieces. I still chase renders past the point where I should have stopped. I still publish things and wish I'd held them. This isn't a discipline you arrive at. It's a discipline you practice.
If you're reading this and you've been sitting on a piece because you're not sure it's ready, take this as your sign. Publish it. Make the next one. The community is here for the work, not the perfection.
I'm glad you're making things.
— Matt