A new study has found that AI-generated pictures of food can trick people into making the same nutritional judgments they would make about real food. But those same images still fall short when it comes to making people actually want to eat what's shown.
The research was published in the journal Scientific Reports. It looked at how people respond to AI-generated food images compared with real, validated food photos.
Researchers recruited 87 participants for the study. Each person rated 60 pairs of images, one AI-generated and one real, across four categories.
Those categories were realism, willingness to eat, calorie content, and healthiness. The goal was to see whether AI images could stand in for real food photos in research settings.
What the Study Found
The results showed a mixed picture. When it came to calorie content and healthiness, ratings for AI images and real images were close enough that there was no clear difference.
But realism and appetite were a different story. Participants rated AI-generated food images as less realistic than real photos. They also reported a lower desire to eat the food shown in AI images.
Researchers also ran a deeper analysis to see what was driving these results. They found that the gap in realism between AI and real images predicted the gap in both healthiness ratings and appetite ratings.
That connection did not hold for calorie content. In other words, how real a food image looked mattered more for whether someone wanted to eat it than for how many calories they thought it contained.
The Hyperrealism Exception
Not every AI image fell short on realism. Out of 60 AI images, 14 were rated as more realistic than their real photo counterparts.
Among those 14 standout images, 11 also scored higher on willingness to eat compared with the real version. This suggests that when AI food images do look more convincing than reality, people respond to them more strongly.
This pattern echoes something researchers have seen before with AI-generated human faces. Some AI faces have been rated as more realistic than actual photos of real people, a phenomenon known as AI hyperrealism.
However, the food images in this study did not follow that same pattern overall. On average, AI food pictures were seen as less realistic than real ones, not more.
Researchers pointed to a concept from cognitive psychology called face space to explain the difference. This theory suggests the human brain has a mental map for recognizing faces, and AI systems may exploit that structure to appear extra convincing.
No similar structure has been confirmed for how the brain processes food images. That could explain why AI hyperrealism did not carry over from faces to food in this research.
The study authors noted some limits to their work. The AI images used were curated ahead of time, with flawed or artifact-heavy images removed before testing.
That means the real-world gap in realism between AI and real food photos could be larger than what this study measured. All images were also shown against plain gray backgrounds, which may not reflect how these photos appear on menus or delivery apps.
The researchers said the findings could matter for fields that rely on food imagery, including clinical research on eating behavior and appetite. They also pointed to possible uses in food delivery apps and grocery platforms, where realistic AI images could shape what people think about a product's healthiness or appeal.
