What changed
Based on reporting by The Verge, restaurants, cafes and brands are increasingly using AI-generated food imagery, sometimes producing worms, holes, cracks and malformed dishes instead of appetising promotions. Chris Russell of the University of Oxford and Giovanbattista Califano of the University of Naples Federico II explain that diffusion image models build from noise, often locking in a bad coarse structure before adding detailed texture.
Why This Matters
Our view: this is a practical warning against treating image generation as a hands-off production shortcut. Food is unusually unforgiving. A plausible plate has to make physical sense at a glance: noodles must end somewhere, seeds must stay on the bun, and ice cream should not resemble building material.
The report’s explanation points to a useful dividing line. These are not merely cosmetic blemishes that a sharper prompt will always scrub away. When the underlying shape is wrong, adding glossy texture can make the result more vivid and more revolting at the same time. For any product where the image itself carries the promise, human review is not optional.
Impact assessment
Restaurants, cafes and food brands using generated visuals are immediately exposed to promotions that can look unappetising when uncontrolled strands, holes and repeating textures spill beyond sensible food boundaries. The commercial harm is not established in the report, but the quality risk is visible before an image ever reaches a customer.
Comments
No comments yet.