The sameness problem behind those unappetizing AI-generated menus
TechCrunch Amanda Silberling
AI menus are making food look weirdly perfect, and people can tell. The same polished look keeps showing up because these models copy the same narrow style.
Based on reporting by TechCrunch, Amanda Silberling — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The odd thing about AI-made menus is that they can look wrong even when nothing is obviously broken. A bagel sandwich illustration might be smooth, symmetrical, and too neat to trust. A burrito might be so aggressively melty it stops looking like lunch and starts looking like a test of patience. That uneasy feeling is the point: the images are often trained toward a narrow idea of what “pleasing” food should look like.
Reality Defender CTO Alex Lisle says the models behind tools like ChatGPT and Midjourney learn by spotting patterns in huge piles of data, then guessing what a user wants next. If a model is asked for a burger menu, it may lean on the same sort of chain-restaurant material over and over again. That helps explain why the results can feel familiar in the worst possible way. The output keeps drifting toward the same rounded buns, the same perfect scoops, the same glossy sameness.
There’s also a feedback loop problem. Lisle said model collapse happens when a model keeps eating its own output until the system degrades, though he says what’s happening with menus is more like convergence. The quality doesn’t fall off a cliff. It just gets smoother, flatter, and more samey. And if AI-made menus get edited repeatedly — changing prices, swapping item names, tweaking details — the food images can become even more polished and less believable each time.
Lee Rainie, who directs the Imagining the Digital Future Center at Elon University, put it plainly: these systems are optimized for pleasingness and for not offending anyone, and that can turn into homogenization. People notice it, even if they can’t fully explain what they’re seeing. Research from the University of Duisburg-Essen in Germany found that near-real AI food images can trigger an uncanny valley effect, making people more uneasy than pictures that are obviously fake. That’s a bad sign for restaurants, but it’s also a bigger warning about how much we now rely on images and video to mean something is real.
My take — AI-written commentary, not fact-checked reporting
The real problem isn’t that AI menus are ugly; it’s that they’re bland in a way that feels dishonest. That’s the same trap running through a lot of AI right now: optimize for “pleasing” long enough and you end up with mush, not quality. The food may be fake, but the monotony is very real.
Read more about this at: TechCrunch
