Every ugly AI poster you’ve seen on a coffee shop bulletin board was made by a person who didn’t care. That’s the part of this conversation nobody wants to sit with. We’ve spent months blaming the model, the prompt box, the pixel-smeared hands, the garbled text where a phone number should be. But tools don’t have taste. People do. And the flood of what one writer aptly called a “ChatGPT flyer pandemic” isn’t evidence that image generators can’t make a decent poster. It’s evidence that a lot of people making posters never learned what a decent poster looks like.
I review AI toolkits for a living, which means I spend an unreasonable amount of time watching people use software badly and then blame the software. This is the clearest case I’ve run into yet.
What the ugly flyers actually tell us
The examples circulating online — the ones readers keep sending in to publications collecting the worst offenders — share a family resemblance. Text that dissolves into pseudo-letters. Lighting that belongs to three different times of day. Four fingers on one hand, six on the other. A general plasticky sheen over everything, like the whole image was dipped in wax.
Those are all fixable. Not theoretically fixable — fixable in the sense that a working designer with an afternoon and a layout tool fixes them routinely. The AI output was never the finished poster. Somebody decided it was.
That decision is the actual failure point. The generator produced a draft and the human accepted it. No cropping, no typographic pass, no moment of stepping back and asking whether the thing communicates the event date legibly to someone walking past at three miles an hour. The tool did its job. The person skipped theirs.
The contest problem is different, and worth separating
When an AI-assisted poster won the Ohio State Fair’s poster contest, the backlash wasn’t really an aesthetic argument, even though the winning entry reportedly contained the usual generated-image errors. It was a fairness argument. Contests are about attribution and effort, and people felt like the rules hadn’t caught up to the entries.
I think that’s legitimate, and I also think it’s a separate conversation from whether the poster looked good. Conflating the two muddies both. You can hold the position that AI entries should be disclosed or barred from human design competitions while also admitting that generated imagery, in hands that know what to do with it, can produce something worth hanging on a wall.
The contest story got traction because the errors were visible. If the winning entry had been cleaned up properly, nobody would have noticed, and the ethics question would have gone unasked. That should tell you something about how much of the current outrage is actually about craft quality versus process.
Designers are already proving the point
The most useful development in all of this is the human designers now taking these generated messes and reworking them. It’s become a small genre: take a terrible flyer, rebuild it, post the before and after. The results are often striking, and they demonstrate something the “AI versus humans” framing keeps obscuring.
The redesigns aren’t wins for humans over machines. They’re wins for people who understand hierarchy, contrast, spacing, and restraint over people who don’t. The designer fixing an AI poster and the designer fixing a clip-art poster from 2004 are using identical skills. The source of the raw material changed. The judgment required didn’t.
What this means if you’re evaluating tools
A few things I’ve landed on after watching this play out:
- Judge a generator by its ceiling with a skilled operator, not its floor with an unskilled one. Every tool looks bad in the wrong hands.
- Treat generated images as raw assets. If your workflow ends at the download button, your workflow is incomplete.
- Handle type separately. Text rendering is still where these models fall apart most visibly, and there’s no reason to fight the model when a proper layout tool solves it in minutes.
- Build in a review step, even if the reviewer is just you with fresh eyes an hour later.
The uncomfortable truth for the AI-optimistic crowd is that these tools don’t remove the need for design sense — they make the absence of it much more obvious, much faster, at much greater volume. The uncomfortable truth for the AI-skeptical crowd is that the posters getting mocked aren’t proof the technology can’t work. They’re proof that speed without judgment produces garbage, which was true long before any of this.
Good posters were always rare. We just have a faster way to make bad ones now, and a clearer view of who was relying on friction to hide their lack of taste.
🕒 Published: