Why AI food looks like that

AI-generated food images frequently appear grotesque and unappetizing, featuring items like "donut shrimp" and "wormlike noodles." This is primarily due to the limitations of diffusion models, which start with noise and struggle to generate fine, terminating structures like noodles or consistent textures, leading to unnatural patterns and "model collapse" when trained on other AI outputs. Furthermore, AI has no conceptual understanding of food; it merely reproduces statistical visual patterns without grasping context, resulting in textures that might look normal in architecture but become horrifying on food. This triggers a visceral human disgust response, as we are evolutionarily tuned to detect signs of spoilage or contamination in food, making the "uncanny valley" for AI food particularly unsettling.

(Source：The Verge)