I spent my first six months with AI image tools making nothing worth keeping.
Not because I couldn’t prompt. I got good at that fast. I could summon any style, any lighting setup, any era. Baroque chiaroscuro? Done. 1980s editorial with cross-processed film grain? Two tries. The machine never said no.
That was the problem.
When the tool never resists you, you stop asking yourself the only question that matters: is this actually worth making?
Here’s something strange: the easier the tool gets, the harder the work becomes.
When I started with AI visuals, I thought the challenge was technical. Learn the syntax. Master the parameters. Stack the right modifiers. And for a while, that felt like progress. Every week I added a new trick. I could make the machine do things that would take a photographer weeks of prep.
But my output was empty.
Not ugly. Not broken. Just hollow. Images that looked like they belonged in a pitch deck for a startup that doesn’t exist. The kind of work that gets a “nice” and then gets scrolled past.
The paradox: when the tool removes every obstacle, you’re left alone with your taste. And most people discover they haven’t built any.
The real bottleneck in AI visuals isn’t prompt engineering. It’s having a point of view. Most people using these tools can’t answer three basic questions:
1. What feeling am I trying to produce?
Not “what style.” Not “what lighting.” What feeling. Is it dread? Longing? The specific loneliness of a hotel lobby at 2 a.m.? If you can’t name the feeling, the AI will default to the most generic emotional register available: mild awe. Everything becomes a sunset. Everything becomes epic but weightless.
2. Where do I want the viewer’s eye to land first?
This is composition, but older than composition. It’s the director’s instinct: you have two seconds before the viewer decides whether to keep looking. Where do you put them? The AI, left to its own devices, puts them in the center of a symmetrical frame, looking at something glowing. That’s not a choice. That’s a surrender.
3. What am I willing to leave ugly?
This is the question that separates you from the future from prompters. Pro-grade images have tension. They have elements that feel unresolved, accidental, human. The light bleeding where it shouldn’t. The motion blur on the one figure who turned too fast. The artifact you noticed in iteration #14 and decided to keep because it made the image feel alive. Beginners remove every flaw. Professionals know which flaws belong.
Without answers to those three questions, you get what I call the Default Aesthetic.
Symmetrical compositions with a glowing focal point dead center. Skin that looks like polished resin. Lighting that quotes golden hour without understanding why golden hour works — the warmth is there but the directional weight isn’t, so everything looks backlit and weightless. High contrast applied as a vibe, not as a storytelling tool.
It looks expensive. It means nothing.
I see this aesthetic everywhere: on LinkedIn carousels, on AI art Instagram accounts, in pitch decks for products that haven’t shipped. It’s the visual equivalent of corporate ambient music. Pleasant. Forgettable. Designed by nobody for nobody.
And here’s the uncomfortable part: the AI isn’t doing this to you. You’re doing it to yourself. The Default Aesthetic is what you get when you outsource every decision to the machine. Style without substance. Production value without point of view
So how do you build taste?
First, let’s kill the idea that taste is something you’re born with. It isn’t. Taste is pattern recognition built through repeated, deliberate looking. It’s the residue of thousands of decisions about what works and what doesn’t.
Here are three exercises I’ve used with clients — and with myself:
The Reverse Gallery. Take 20 AI-generated images you made in the last month. Print them or lay them out full-screen. Now remove every one that could have been made by someone else. Not “is it good?” but “is it mine?” You’ll be left with maybe three. Study those three. What do they share? That’s the beginning of your visual voice.
The Single-Frame Study. Pick one photograph, painting, or film still that genuinely moves you. Not one you admire — one that hits you in the body. Now spend 30 minutes with it. Trace where your eye moves. Identify every light source. Note what’s in shadow and what isn’t, and ask why. Write down the feeling in one word, then in one sentence. This is what taste work looks like. It’s slow. It’s unglamorous. It works.
The Ugly Element Exercise. For your next project, decide before you start what you will not fix. One thing. A blown-out highlight. An awkward crop. An AI artifact that looks almost intentional. Ship it with that flaw intact. See if anyone notices. More importantly, see how it changes the image’s emotional weight.
The machine can make anything. That fact is not going away.
What changes is what you bring to it. Not better prompts. Not more parameters. A clearer sense of what you’re trying to say and the discipline to stop when you’ve said it.
That’s the work. It happens before you open the tool. It happens in galleries, in film stills, in the way you notice light falling on a wall while you’re waiting for coffee. It happens in the thousands of micro-decisions where you choose what to keep and what to let go.
The AI will always be faster than you. Let it be. Your job is to know what’s worth being fast about.
This is exactly the shift I guide in my e-book “Turn Your AI Images Into High-End Visuals.” It’s not a collection of prompts. These are the guidelines in changing how you think about AI images — from “how do I make this?” to “why should this exist?” If your images look good but feel empty, this is for you.
And if you want to go deeper: I work 1:1 with creators, brands, and founders on visual strategy and AI-powered branding. Not prompt coaching. The full thing. Visual identity, art direction, building a language that’s actually yours. If that sounds like what you need — reach out.
Thanks for today.
See you next time.
Best, Pawel
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.