The sameness problem behind those unappetizing AI-generated menus
There's a quiet but growing tension in the restaurant industry: owners are reaching for generative AI tools to produce polished food visuals quickly and cheaply, while customers are increasingly walking in with expectations shaped by images that don't reflect what actually arrives at the table. The mismatch isn't always conscious, but it's felt.
The core issue is what critics and designers have started calling the "sameness problem." Generative image models are trained on vast datasets of existing photography, and they tend to produce outputs that converge on a kind of averaged visual ideal - perfectly lit, flawlessly composed, and thoroughly generic. A burger looks like every burger; a pasta dish could belong to any restaurant on any continent. That homogeneity strips food imagery of the specificity and authenticity that once made a well-shot menu photo trustworthy.
For diners, the disconnect registers somewhere between intuition and disappointment. Food photography has always involved a degree of idealization - professional shoots use styling tricks that no kitchen replicates exactly - but there's a difference between a touched-up photograph of a real dish and a fully synthetic image that bears no direct relationship to anything on the premises. Customers who grew up visually literate in the age of food blogs and Instagram have developed a sensitivity to cues of authenticity, and AI-generated visuals often fail those cues in ways that are hard to pin down but easy to feel.
The practical consequences extend beyond aesthetics. When the food on the plate diverges too sharply from the image that sold it, trust erodes - and in a sector where repeat business and word of mouth are critical, that erosion matters. The appeal of AI image generation for small restaurant operators is understandable: professional food photography is expensive and time-consuming. But the shortcut may carry a hidden cost, and as AI-generated imagery becomes more prevalent, customers are likely to grow more attuned to it, not less. Authenticity - the specific, imperfect, real thing - remains difficult to automate.
