Why AI food looks like that

There's a specific kind of horror in watching a diffusion model try to render a Reuben sandwich. It produces something that looks like it was dredged from a deep-sea trench, garnish and all. The Verge's catalog of AI food failures is funny at first glance, but it's actually a precise diagnostic of what large generative models misunderstand about the physical world. Food is uniquely punishing terrain for these systems because a food image carries enormous multimodal priors — we know what bread should feel like, what a noodle should bend like, what a fried shrimp's crust should crack like. When a model averages millions of training images, it produces the "platonic" food, which is to say, food stripped of all the specific, local material properties that make it appetizing. It renders the statistical center of every donut and every shrimp ever photographed, and that center is a nightmare. The deeper story here is economic. Restaurants and cafes are adopting AI imagery not because it's good but because it's cheap, and the cost structure of marketing has shifted so dramatically that paying a photographer or food stylist now looks like a luxury. This is the classic "slop pipeline" — generative AI is being deployed in contexts where the quality bar is low enough that "good enough to trick a scrolling thumb" is the actual business requirement. The result is a visual race to the bottom where the aesthetic standard for food advertising is being set by whatever garbage the cheapest model produces. What makes the AI food phenomenon more than a meme is the feedback loop it portends. These AI-generated images are already contaminating search results, delivery apps, and social feeds, which means they will inevitably be scraped into the next generation of training data. The models will then be trained on their own output, reinforcing the smooth, glossy, structurally impossible aesthetic. We're watching the formation of a closed loop of visual degradation in real time — the uncanny valley is being industrialized and made permanent because the economics reward it. The encouraging counter-signal is that people notice. The visceral disgust response to AI food is stronger than the response to generic AI art, precisely because food is so deeply tied to bodily expectation. That disgust is a form of resistance, but it's also a warning label: any industry that replaces authenticity with statistical approximation is gambling that viewers won't care. For now, at least, they care — they care enough that a donut with shrimp legs becomes a story people share with revulsion. Whether that revulsion translates into consumer behavior is another question, but it's the only meaningful check on the slop pipeline. Source: https://www.theverge.com/ai-artificial-intelligence/989376/ai-generated-food-why-does-it-look-like-that