Why AI-generated restaurant menus all look the same

Restaurant menus generated with AI image tools have developed a distinctly uncanny look — impossibly symmetrical, unnaturally smooth illustrations that feel wrong even when it’s hard to say why. “It’s almost like an alien trying to make a pizza without understanding its core principles,” Alex Lisle, chief technology officer at AI-detection startup Reality Defender, told TechCrunch.

Lisle said the pattern traces back to how large language models and diffusion models are trained: on vast datasets from which they learn to predict what a user wants when asked to generate something like a burger-restaurant menu. Much of that training data resembles the same narrow slice of chain-restaurant imagery — Lisle compared the look to a Chili’s menu from 2015 — so AI outputs tend to reproduce and reinforce that same aesthetic. When AI-generated menus end up back in future training data, the effect compounds, a phenomenon Lisle described as “convergence,” distinct from the more extreme failure mode known as model collapse, in which a model trained too heavily on its own outputs degrades until it becomes unusable.

Lee Rainie, director of the Imagining the Digital Future Center at Elon University, said the underlying training optimizes for images that are “pleasing” and inoffensive, which tends toward homogenization: AI, he said, is known to shave off the edges in both images and language. On X, a user named Labtec showed how repeatedly editing an AI-generated menu — tweaking prices or item names roughly 100 times — made the food look progressively more rounded and artificial with each pass; TechCrunch replicated the experiment and found similar results, with Labtec writing that the end result was genuinely uncomfortable to look at.

Researchers at the University of Duisburg-Essen in Germany found that AI-generated food images can trigger an “uncanny valley” effect, in which images that look almost — but not quite — real provoke more disgust and unease than images that are obviously fake, a reaction Rainie said intensifies given growing public wariness of AI content generally.

Why it matters

As AI-generated visuals become harder to distinguish yet still trigger an instinctive unease, businesses that lean on them risk alienating customers precisely because the images look almost, but not quite, real.