The Texture Trap: How AI Is Making Food Look More Detailed, But Less Appetising

A few days ago I put two burgers on our Instagram story and asked one question. Which one would you order. One of them I had shot myself, lights set, bun toasted, sauce placed, the whole afternoon spent on it. The other one was generated by AI and, if I am honest, it looked cleaner than mine. Sharper edges, better symmetry, a sesame pattern so even it could have been printed. The poll came back and most of the people who voted chose the real one.

It was a small poll on our own page, so I am not going to pretend it is research. But it confirmed something I have been watching happen for the last year, and it is worth talking about properly, because more and more restaurants in Dubai are quietly replacing their menu photography with generated images and nobody is really asking what that does to the person holding the phone.

Photography has always been a form of proof

Before we talk about AI at all, we need to be clear about why food photography

exists. It is not decoration. A photograph is evidence that a human being stood in front of something and saw it. That is the whole contract. When you look at a picture of a bowl of ramen or an American steak burger, what you are really reading is a message from another person saying I was here, this is what it looked like, and this is what you are going to get.

That matters more with food than with almost anything else, because the viewer already has the memory in their body. You have eaten a burger before. You know how the bun gives way, how the pickle cuts through the fat, how the cheese behaves when it is warm. You do not consciously think about any of this. You look at the photo and your mouth decides. My job as a photographer is to give your mouth something true to decide on. However messy the burger is, however uneven the cheese sits, however awkward the drip, that is the real burger, and the mess is not a flaw in the photograph. The mess is the evidence.

Then that experience travels. Someone eats something good and immediately wants to show it to other people. The photo is how they say I had this and you should have it too. The entire food economy on our phones runs on that one human instinct.

Where AI quietly breaks the contract

An AI image has no plate behind it. Nobody stood in front of anything. The model has looked at a very large number of burgers and produced something that resembles the idea of a burger, and that is a completely different object from the burger your kitchen actually serves. There is no shoot, no chef, no specific Tuesday afternoon when that patty existed. The proof is missing, and the viewer feels the absence even when they cannot name it.

What they can name, usually, is that it looks too perfect. The bun is flawless. The pickle sits exactly where a pickle should sit. The patty has no collapsed edge, the lettuce has no bruise, the sauce has no accident in it. There is no burger like that anywhere in the world, and some part of the customer knows it. The moment a picture reads as too good to be true, the next thought is automatic. This is not what will arrive. That thought is expensive, because it happens a second before the order button.

Researchers in robotics gave this feeling a name long before any of this software existed. The uncanny valley describes what happens when something comes very close to real without getting there, and how quickly familiarity turns into unease. Synthetic food imagery has walked straight into that valley, and it has done it by being better, not worse.

The texture trap

Here is the strange part, and it is the reason for the title. The output is not getting cruder. It is getting more detailed, and the detail is exactly what is going wrong. Generated images have become extremely good at surface. Crumb, char, condensation, sesame, the fine grain of a toasted bun. What they are not good at is the irregularity underneath that surface.

So you end up with a bun where the seeds are distributed with a mathematical evenness no baker could produce, sitting in clusters that repeat across the frame. Human eyes are genuinely sensitive to dense repeated patterns. There is a well documented discomfort response to clustered patterns, and while nobody expects a burger to trigger it, the mechanism is close enough that the image starts to feel wrong in a physical way rather than an intellectual one. My colleague described it as disgust before judgement, and that is accurate. You do not decide you dislike the image. Your stomach decides first.

Real food does not do this. Real sesame falls where it falls. Real char is heavier on one side because the grill is hotter there. Real cheese pulls unevenly because gravity is involved. Every one of those irregularities is a small signal of authenticity, and when you remove all of them in pursuit of detail, you remove appetite along with them.

What a photograph carries that a render cannot

I want to be clear that I am not against this technology. I am very much pro AI. I use it where it saves time on repetitive work and I would be happy if it did more. But food photography is not only about the food. It is about the experience around the food.

A hand holding a burger tells you how big it is. A person smiling mid bite tells you it was worth it. The cheese actually pulling, the drip actually running down a thumb, the steam actually rising off loaded fries while somebody reaches in. These are not decorative choices, they are testimony. Even if a model can mimic all of them perfectly, there is no personal experience sitting behind the image. The sentence food photography is really saying is this happened to me, and if you eat it, it will happen to you too. That sentence requires a witness. That is the part that cannot be generated.

Every shoot is a disaster, and that is the job

People think photographers are hired for equipment. They are not. A photographer or any artist is not working at their full calibre until they are a good disaster manager. Most photographers understand photography reasonably well at some point in their career. What separates good from great is how they handle the moment it all goes wrong, and the bigger the disaster, the better the final picture usually is, because the fix forces a decision nobody would have planned.

Shoot gelato and you will learn this immediately. You spend twenty or thirty minutes building one frame and the scoop surrenders in under a minute. You rebuild and go again. Shoot burgers and everything is balanced on the edge of collapse, and then it collapses, and you rebuild it better than it was the first time, because by then you know exactly what was going to fail. That knowledge is the product. It is not in the camera.

This is also my answer to the clients who tell me they can do it on an iPhone, or now, that they can do it with AI. I never argue. You are not hiring me for a Nikon. You are hiring me because I know where the light has to come from, which angle hides the imperfection and which angle uses it, how to edit it so it still looks like food and not like plastic. A chef can cook the most incredible ramen. I can make ramen look beautiful even when it is not cooked. Those are two different skills and both of them are earned over years. I let people go and try it themselves. The work speaks later.

The only question a restaurant owner has to answer

When I saw that poll result, what I actually wanted to do was ask every restaurant owner in the UAE one question. Do you want a good looking picture of your dish, or do you want orders.

Everyone answers orders, every single time. So follow that answer through. An order happens when somebody believes the plate in the photograph is the plate that will arrive at their door. Belief comes from proof. Proof comes from someone having actually been in the room with that food. And when the plate does arrive and it matches, the customer photographs it themselves, sends it to someone else, and the cycle starts again on your behalf. That loop is worth far more than one flawless render.

Break the loop and the cost is not just a lost order. It is a refund request, a one star review about the picture being nothing like the product, and a customer who will never fully trust your menu again. Most delivery platforms already expect menu imagery to represent what is actually served, and advertising standards generally take a dim view of food photography that misleads about the product itself. Generated imagery sits in uncomfortable territory there, and I think that conversation is coming.

Where I think this lands

AI will keep getting better at surface and it will keep being useful. I expect to use it more, not less, for the repetitive parts of this work. But a menu is not a mood board. A menu is a promise, and a promise needs a witness behind it.

That is why I asked the question on Instagram instead of writing a lecture, and it is why we are taking this conversation further. If you have an opinion about it, especially if you run a kitchen, I want to hear it. The customer has already told us what they prefer, and they did it without knowing which image was which.