A Letter to Myself, Written the Week I Almost Quit Using AI Image Tools Entirely

It's the tail end of March. The heating in my parents' house is still running even though the afternoons have started to feel like something other than winter, and I'm sitting at the old wooden desk in my childhood bedroom — the one with the drawer that sticks, the one where I used to do homework I didn't understand — with my laptop open and a feeling I can't quite name sitting somewhere between my sternum and my throat.

I've been back here for two weeks. A short stay that keeps extending. You know how that goes.

I want to write this down while it's still uncomfortable, because I think the discomfort is the point.

What Happened

A few days ago, I was scrolling through a design community forum — the kind of low-key place where people post work-in-progress and ask honest questions — and I came across a portrait series someone had made using an art photo generator. The images were extraordinary. Not in a technically-perfect way. In the way that makes you stop and forget what you were looking for.

The faces were rendered with a specificity that felt almost biographical. The hands — and I noticed the hands specifically, because hands are where AI image generation tends to fall apart, where fingers multiply or joints bend the wrong way or the whole structure goes soft and unconvincing — the hands in these images were right. Not photographic-right. Right in the way a painted portrait is right: slightly stylized, but structurally coherent, emotionally legible.

I sat with those images for longer than I'd like to admit.

And then I opened my own recent work folder and felt something I'm going to call, honestly, inadequacy. Not because my work was bad. But because I couldn't tell, looking at it, whether the good parts were mine.

The Question I Keep Avoiding

Here's the thing I've been circling for months without landing on:

When I use an AI photo generator to produce an image — when I write the prompt, iterate on it, select from the outputs, do the post-processing, make the compositional decisions — how much of that is me?

I know the theoretical answer. I've read the essays. Curation is authorship. The prompt is the brush. The eye that selects is the eye that creates. I believe these things, mostly, in the abstract.

But then I look at those hands in that portrait series, and I think: I didn't make that. I couldn't have made that. Whatever process produced those hands — the model's training, the weights, the millions of images it learned from — that's not my skill. I just typed words into a box.

But here's where it gets tangled: the person who made that portrait series also typed words into a box. And their results were different from mine. So something is different between us. Is that something "skill"? Is it taste? Is it just luck and iteration count?

I don't know. I genuinely don't know, and I'm suspicious of anyone who answers too quickly.

The Specific Thing That Broke Me (A Little)

I've been working on a personal project for about four months. A series of portraits — imagined people, no references, built entirely through prompting — that I wanted to feel like documentary photography from a country that doesn't exist.

The faces came relatively easily. Faces are something current AI image tools handle with a confidence that can tip into uncanny, but with the right prompting you can pull them back toward something human and specific.

The hands kept failing.

Not always. Maybe one in eight generations produced hands that felt structurally honest. The rest ranged from slightly wrong to actively disturbing — extra joints, merged fingers, a kind of boneless softness that reads immediately as artificial. I tried every approach I'd read about: close-crop prompts, explicit anatomical descriptors, reference image uploads, negative prompts listing every hand pathology I could name.

The hit rate improved. It never got reliable.

I ended up cropping most of the hands out of the final series. Which is a creative choice, technically. But it's also a workaround. And I'm not entirely sure which one it is.

When I saw that forum post — those hands that worked — my first feeling wasn't inspiration. It was something closer to: what am I missing, and is it something I can learn, or is it something I just don't have?

What I Actually Think (Today, Anyway)

I've been using Photogenerator and a few other tools long enough to have an opinion, even if the opinion keeps changing.

Here's where I am right now, sitting at this sticky-drawered desk with the afternoon light coming through a window I haven't looked out of in years:

The tools are genuinely capable of producing images I couldn't produce any other way. That's real. That's not nothing.

But the capability isn't evenly distributed. Faces: yes, mostly. Environments and textures: yes, often beautifully. Hands, feet, complex structural anatomy, anything that requires the model to understand how a body occupies space and exerts force: inconsistently, unpredictably, sometimes not at all.

And the inconsistency matters, because it means I can't fully trust the output. I can't look at a generated image and know whether the hands are right or whether I've just gotten used to them being wrong. My eye calibrates to what I see most often. That's a problem.

But here's the other side of it, the side I keep arguing with myself about: every medium has structural limitations. Film photography has grain and exposure constraints. Illustration has the limits of the illustrator's hand. The question isn't whether the tool has limitations. The question is whether the limitations are generative — whether they push you toward something you wouldn't have found otherwise — or whether they're just walls.

Cropping out the hands was a wall. But the country-that-doesn't-exist series exists because of the tool. Both things are true.

The Part I'm Writing for Future Me

I don't know if you — future me, reading this in some other season, at some other desk — kept going with this. I don't know if the hand problem got solved by better models, or better prompting, or whether you just made peace with the crop.

What I want you to remember is the feeling of sitting with those images on the forum and feeling inadequate, and then — after a while, after the feeling settled — feeling something else. Something more like: okay, so that's possible. What does that mean for what I'm doing?

Not competition. Not comparison. Just: the horizon moved, and that's information.

The tools will keep changing. The limitations will shift. Some walls will become doors and some doors will become different walls. That's just how this works.

How This Ends (For Now)

It's late afternoon now. The light through the window has gone from white to yellow to the specific amber that only happens in early spring, when the sun is still low but the air has stopped being cold. The desk lamp isn't on yet. I'm working in that in-between light where everything looks slightly more serious than it is.

I have the portrait series open in one window. The hands are cropped. The faces are specific and strange and, I think, true to something.

I'm going to keep going.

Not because I've resolved anything. But because the alternative — stopping, waiting until I understand what I'm doing — has never once worked for me in any medium, with any tool, at any desk.

The amber light is almost gone. I should turn the lamp on.