Updated
Updated · The Verge · Sep 4
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.

Insights

Could the restaurant industry's quest for the perfect menu image actually destroy AI models through a digital feedback loop?
If traditional food ads already use fake ingredients, why do perfectly smooth AI-generated burgers trigger our primal disgust reflexes?
When restaurants use AI to fabricate menus, who is held responsible when the real dish fails to match the fantasy?