TechCrunch has reported on a growing tension in the restaurant industry, where proprietors are turning to generative AI tools to produce menu photography and descriptive copy, only to find that customers recoil from the results in ways they often cannot quite articulate. The piece identifies a kind of instinctive revulsion — something uncanny and off — that diners experience when confronted with AI-produced food imagery, even when they cannot name the technology responsible.
To understand why this matters, it helps to trace how food presentation became so freighted with meaning in the first place. Restaurant menus, particularly their visual elements, have never been straightforward documentation. A photograph of a dish is a promise, a piece of theater, and a trust signal all at once. The food styling industry exists precisely because making real food look appealing on camera is extraordinarily difficult and expensive — hence the legendary tricks of the trade, from motor oil standing in for syrup to raw meat treated with chemicals to hold its color under hot lights. What human stylists have labored for decades to perfect is a kind of controlled illusion that nonetheless reads as real to the viewer's eye. Generative AI, trained on vast pools of images, learns the statistical average of what appealing food looks like, and that is precisely where the problem begins.
What AI image generation tends to produce is something like a composite of every burger, every pasta dish, every plated dessert it has ever encountered. The result is technically accomplished in surface terms — colors are vivid, textures are present — but the food exists in a kind of hyperreal nowhere. Portion sizes drift into the implausible. Shadows fall from no coherent light source. Ingredients appear in combinations that would be physically impossible to plate. And crucially, the food looks like no specific dish that any specific kitchen could actually produce. It carries the faint smell of nowhere, and customers, this suggests, are picking up on that mismatch at a level below conscious recognition.
This connects to a deeper pattern in how generative AI tools have been deployed across creative industries. The technology optimizes for the average, for the statistically central representation of a category. That works tolerably in some contexts and fails badly in others. Food is one of the worst possible applications, because cuisine is fundamentally about specificity and locality — this chef, this region, this season, this technique. A menu photograph is not just asking the viewer to find food attractive in the abstract; it is making a claim about what will arrive at a particular table. When the image is generic by construction, it cannot make that claim convincingly, and some part of the viewer registers the breach.
The business logic driving restaurant owners toward these tools is easy to understand. Professional food photography is genuinely expensive, requiring a photographer, a food stylist, often a prop stylist, and hours of setup time. For independent operators running on thin margins, the appeal of generating a plausible-looking image in minutes for near zero cost is obvious. The generative AI companies selling these tools have been careful to emphasize the cost and time savings while being less forthcoming about the perceptual and reputational risks downstream.
The consequences are distributed unevenly. Large chains with established brand photography libraries and the budgets to refresh them professionally are largely insulated from this temptation. The operators most likely to reach for AI-generated visuals are small and independent restaurants, and those are also the establishments for which trust and local identity are the primary competitive advantages over corporate competitors. If AI imagery erodes that trust — even subliminally, even in ways customers cannot name — the damage falls hardest on the businesses that can least afford it. There is also a longer-term risk for the AI tools themselves: if generative imagery becomes widely associated with low-quality or deceptive restaurant marketing, the category could acquire a stigma that outlasts its current technical limitations.
For the food service industry specifically, this moment likely accelerates a bifurcation that was already underway. Establishments that lean into authenticity — real photography, behind-the-scenes content, visible kitchens — may find that the contrast with AI-generated competitors actually sharpens their appeal. The uncanny quality of AI imagery could, paradoxically, make honest documentation of real food more valuable, not less.
What to watch for next is whether platform-level actors enter this conversation. Delivery apps and review sites that host restaurant imagery have enormous leverage over how menus are presented to consumers, and some are already experimenting with disclosure requirements or quality filters. Regulatory interest in AI-generated commercial imagery, while still nascent in most jurisdictions, is another variable worth tracking. And on the technical side, it remains to be seen whether future model generations can produce food imagery specific enough to the actual dishes being served to close the credibility gap — though the structural problem, that training on averages produces averages, will not be easy to engineer away.
