Why AI-generated food images look so unappetizing
Restaurants and brands are increasingly using AI-generated images to promote food, but the results are often disturbing rather than appetizing. Experts explain the technical and psychological reasons behind this.
A growing number of restaurants, cafes, and brands are turning to AI-generated images to promote their food, but the results are frequently unsettling rather than appetizing — think warped, worm-like noodles, holes clustered together, and burgers or ice cream that resemble concrete or rock.
How image generation goes wrong
Most leading image generators rely on diffusion, a process that starts from random noise and gradually removes it to build the requested image. Coarse structures form first, with fine texture details added later. Oxford University professor Chris Russell explains that errors often begin early on, when a model gets an object's basic structure wrong before layering vivid textures on top — the same kind of failure seen when a generated person has six fingers instead of five.
Giovanbattista Califano, a behavioral scientist at the University of Naples Federico II, adds that diffusion models struggle particularly with thin, continuous structures like noodles or strands, which can bleed into parts of the image where they don't belong. Repeating textures such as bubbles or seeds face a similar issue, often spilling beyond sensible boundaries.
AI doesn't actually understand food
Roland Meyer, a professor at the University of Zurich, notes that AI image generators reproduce the look of things without real understanding of the world — a model learns statistically what food tends to look like, but not why. Michael Cook of King's College London adds that this can produce textures that would look normal in an architectural context but become disturbing once imagined as edible.
The training data compounds the problem. Ordinary, "boring" food photos are rarely posted online, while strange or bizarre images tend to spread widely as memes. Models are also increasingly trained on AI-generated material itself, which researchers say can trigger "model collapse" — a growing sameness and degradation in image quality over time.
Scientists note that humans are especially attuned to spotting food that looks wrong, since disgust likely evolved as protection against spoiled or unsafe food — which helps explain why AI-generated food, with its odd textures and clustered holes, tends to feel so viscerally off-putting.
