Why Do AI-Generated Food Menus Look So Off-Putting?

You may have had this experience without realizing why: you walk into a café and browse the menu, only to notice that the sandwich photos look a little too perfect, oddly symmetrical and unnaturally smooth, leaving you with a faint sense that something isn't quite right. You're not being paranoid—you're looking at menus designed by generative AI, a phenomenon that has recently begun to invade the restaurant world.

These images range from the blatantly fake—like a burrito with melting cheese that looks more like abstract art than lunch—to the mundane enough that you only spot the flaw after a careful look. One tech expert described these results as being "like an alien trying to make a pizza without understanding its basic principles."

Why Do All These Images Look Alike?

The heart of the problem lies in how AI models are built and trained. Large language models and diffusion models—the technologies powering chatbots and image generators like ChatGPT and Midjourney—are trained on enormous amounts of data, then extract patterns from it to predict what a user wants when they request, for example, "design a menu for a burger restaurant."

The result is that many of these designs look identical in style, since the models rely on popular restaurant menus that already share similar aesthetics. Every scoop of ice cream is perfectly round, and shrimp look as if they've been genetically modified to devour their own tails, producing what could be called bizarre "food horrors."

The Convergence Phenomenon and the Risk of Model Collapse

When AI models are trained on an excessive amount of content produced by other models, two interrelated problems arise:

• Model collapse: This situation resembles mad cow disease; when a model's outputs are repeatedly fed back into itself, this "inbreeding" becomes destructive until the entire system collapses.
• Convergence: This is less severe, as output quality gradually degrades without becoming completely useless, so images grow more alike and formulaic with each training cycle.

So when a user requests a menu design for a fast-food restaurant, the model draws on menus from famous chains that are already similar, producing a result that mimics the same style, then reinforces it further if this generated menu later returns to the training data.

"Smoothing the Edges" and Aesthetic Standardization

Experts believe these models are optimized to produce what is "pleasing" and "inoffensive," which inevitably leads to homogeneity and standardization. AI is known for "removing the edges" in both images and text, causing content to lose its natural details.

The problem intensifies when restaurant owners repeatedly edit their generated menus to change small details like prices or item names. Experiments on the X platform showed that editing a menu created via ChatGPT a hundred times makes the food look increasingly rounder, smoother, and further from its real appearance with each edit.

Why Do We Feel Repulsed?

This aversion has a scientific basis. Researchers at Germany's University of Duisburg-Essen found that AI-generated food images produce an "uncanny valley" effect, where nearly realistic images provoke a greater degree of disgust and discomfort than clearly fake ones. This feeling is compounded by the cultural context surrounding AI.

Experts note that people have a hard-to-explain sense about artificial content; they "know it when they see it" even if they can't describe why, which explains the intensity of early reactions rejecting restaurants' use of AI menus.

If the public reacts so negatively to these images, that may be reason enough for restaurants to abandon them. But the problem goes beyond the dinner table; AI's ability to standardize what we see raises deeper questions about trust in images, in a world where the rule "seeing is believing" has been fundamental—even in our judicial systems.

✦ بقلم فريق دروب أيديا