Restaurants, cafés and food stalls are increasingly using artificial intelligence to create images for menus and outdoor boards. The pictures can be quicker and cheaper to produce than professional food photography, but some diners say they look unnatural and even unappetising.
Comments on Instagram describe AI-generated food images as repetitive or artificial. Some users say unusual textures remind them of holes or insect nests. Others argue that AI offers small businesses an affordable way to promote their dishes. Germany’s hotel and restaurant association, Dehoga, says cost, speed, time and availability can all make the technology useful to businesses. It notes, however, that there are no reliable figures on how widespread AI-designed menus are in Germany.
A Berlin Lángos stall uses AI because it makes designing a menu easier and faster, an employee told the report. Some customers notice the design, but the employee said their reactions have not been notably positive or negative.
AI food images have also drawn criticism on the Lieferando ordering platform, part of Just Eat Takeaway. Some images circulating online differed substantially from the accompanying descriptions, including a chicken burger shown with rice buns instead of bread. Lieferando said a test of AI-generated images had not worked as intended in some cases. The company apologised for inaccurate pictures, said it was working to remove them and had paused use of the tool. It added that it considers AI potentially helpful to restaurant partners and will continue refining the technology.
A study published in Nature examined how people respond to AI-generated food images. Psychologist and postdoctoral researcher Alexander Diel of the LVR University Hospital Essen and the University of Duisburg-Essen worked with an international team on an online test involving 87 participants. They were shown 60 pairs of images, each containing a real photograph and an AI-generated image of the same food, such as vanilla ice cream or grapes. Participants rated how realistic each image appeared, how willing they would be to eat the food, and how healthy and high in calories they thought it was.
On average, participants considered the AI images less realistic and reported slightly less appetite for the food they showed. Their assessments of the food’s healthiness and calorie content were similar for AI images and real photographs.
Diel suggested that images may seem less appetising when they look too perfect, with overly smooth or symmetrical surfaces, or contain other visual errors. People can also be cautious about unfamiliar-looking food. He said AI images do not reliably prompt the same response as photographs of real food, and that noticeable flaws can make them off-putting.
There is also a difference between an AI-generated picture and the dish a restaurant actually serves: the image may depict a generic meal derived from a model’s training data rather than that restaurant’s product. If customers find that the real dish does not match the menu image, it can lead to frustration. Dehoga described authenticity as central to the promise restaurants make to diners.
Sources: Handelsblatt.