Why AI-generated restaurant menus look so unsettlingly similar
AI-generated food illustrations on restaurant menus are increasingly triggering unease due to their overly smooth, symmetrical look. Experts say the cause lies in how these models are trained on repetitive data, leading to visual homogenization.
More cafes and restaurants are turning to AI-generated illustrations for their menus, but many customers find something off about them — the images look too smooth, too symmetrical, unsettling in a way that's hard to pin down. Sometimes the results are obviously fake, like a burrito with cheese so bubbly it looks more like abstract art than food.
The root cause
Reality Defender CTO Alex Lisle explains that large language models and diffusion models, which power tools like ChatGPT and Midjourney, are trained on massive datasets and try to predict what a user wants to see. Because much of that training data draws from similar existing menus, including those of popular fast-food chains, the outputs trend toward sameness — a phenomenon called convergence. This is distinct from the more severe "model collapse," where training on a model's own generated output eventually renders it nearly useless.
Optimizing for pleasantness breeds uniformity
Lee Rainie, a researcher at Elon University, notes that datasets are optimized to be pleasing and inoffensive, which drives homogenization — AI tends to smooth away distinctive or unusual details. An experiment in which a social media user edited an AI-generated menu 100 times showed the food growing progressively rounder and smoother with each pass, producing an increasingly uncomfortable result.
An uncanny valley effect
Researchers at the University of Duisburg-Essen in Germany found that AI food images that look almost real trigger more disgust and unease than images that are obviously fake — an "uncanny valley" effect. Lisle adds that this reflects a broader shift: it's becoming harder to trust what we see and hear in the digital age.
