Why AI food looks like that

AI-generated images of food often look disturbing and inedible, featuring worms, holes, cracks, stringy textures, and materials that resemble construction materials rather than food. Restaurants, cafes, and brands increasingly use AI to generate promotional food images, sometimes producing results that seem counterproductive to their marketing goals. The problem stems from how diffusion models work: they start with noise and gradually add details, but can get basic structures wrong before adding fine textures on top, and they struggle particularly with thin, continuous structures like noodles and strands that spill into illogical places. Additionally, AI has no actual understanding of what food is or how it should behave in the physical world, only a statistical approximation of how food typically appears, which can result in textures and materials that would be normal in other contexts but become unsettling when applied to food.