Why AI food looks like that
The Verge · 2026-09-04
AI-generated food images often appear unappetizing due to technical weaknesses in image generators, particularly diffusion models. These models begin with noise and gradually remove it, recovering coarse structures first, which means errors in basic object formation can occur early and then be overlaid with vivid textures, leading to issues like six-fingered hands or "donut shrimp." Diffusion models also struggle with generating thin, continuous, and terminating structures, causing "spaghetti-like artifacts" and "noodly tendrils" that lack anatomical or culinary logic, according to Chris Russell, a professor of AI at the University of Oxford, and Giovanbattista Califano, a behavioral scientist at the University of Naples Federico II.
Furthermore, AI lacks a fundamental understanding of what food items like sandwiches or noodles actually are; it only reproduces their visual appearance based on statistical patterns from training data, explained Roland Meyer, a professor at the University of Zurich. This absence of real-world comprehension can result in textures that are normal in other contexts, like architecture, appearing unsettling when applied to food, as noted by Michael Cook, a senior lecturer at King’s College London. The training data itself can be problematic, containing bizarre internet content or even AI-generated material, which can lead to "model collapse" and visual degeneration. Ultimately, humans are biologically programmed to notice when food looks wrong as a protective mechanism, and AI-generated food images frequently trigger primal disgust responses by resembling worms, infestations, or spoilage.