If you treat AI image generation like an exam question, it really exposes someone’s actual skill level. Give the same topic and the same model—one person cranks out a usable image in half an hour, another spends a whole day tweaking and it still looks like cheap plastic. The difference isn’t the tool.
People who know what they’re doing already have the image in their head—they know the lighting, the composition, how big the subject should be, and the prompt is just translating that. People who don’t know just let the model think for them, taking whatever it spits out.
So it’s never about how many prompts you’ve memorized—it’s about your aesthetic sense and visual judgment. The stronger the tool, the more obvious this gap gets, because it shifts the barrier from technique to taste, and taste can’t be rushed.
People who already have a clear image in their head just need to translate it into a prompt. Those who don’t have an image let the model think for them. That’s where the difference lies.
Same prompt, same model, some people finish in half an hour, others take a whole day and still get that plastic look. It really comes down to taste, not memorizing keywords.
I’ve had interns before, and the ones who got work done the fastest were all from an art background—they already had a solid grasp of lighting and composition.