AI Food Images Repel Viewers as Diffusion Flaws Distort Noodles, Holes and Textures
Updated
Updated · The Verge · Sep 4
AI Food Images Repel Viewers as Diffusion Flaws Distort Noodles, Holes and Textures
2 articles · Updated · The Verge · Sep 4
Summary
Restaurants and brands using AI food imagery are ending up with wormlike noodles, hole-riddled burritos and rocklike burgers that look distinctly inedible.
Diffusion models build images from noise, recovering coarse shapes before fine detail, so early structural mistakes get covered with vivid textures instead of corrected.
Thin, continuous forms such as noodles and strands are especially hard for these systems, while bubbles, seeds and repeating patterns often spill beyond sensible boundaries.
The problem is compounded because image models imitate statistical looks rather than understanding what sandwiches, burritos or ice cream are supposed to be in the physical world.
Training on stylized food photography, bizarre internet imagery and even AI-generated outputs can deepen the uncanny, same-looking results already noted in earlier reports.