The sameness problem behind those unappetizing AI-generated menus

AI-generated menus are causing unease among diners due to a shared aesthetic that feels overly perfect and unnatural. These menus are produced by generative AI models that have been trained on a limited selection of aesthetically pleasing data, resulting in illustrations that appear flawless but lack authenticity. For example, the cheese on a burrito might look too bubbly and artistic, rather than realistic.

More commonly, the menus are so ordinary that they only seem off when scrutinized closely. Alex Lisle, Reality Defender CTO, describes the issue as "alien" attempts to create meals without understanding their fundamental principles. The models used to create these menus are trained on vast amounts of data, often incorporating patterns found in previously popular menus, which leads to a repetition and convergence of styles.

This convergence doesn't necessarily result in model collapse but causes the outputs to degrade in quality. Restaurants are unconsciously editing these menus to maintain their appeal, resulting in increasingly smooth and round food images. People often have an instinctive feeling that something is off about AI-generated food images, a phenomenon known as "uncanny valley."

This aversion is further amplified by cultural context, which makes people uneasy about these images, even if they can't fully articulate why. The implications of this phenomenon extend beyond the dining experience and touch on the fundamental changes in evidence perception in modern society.