Summary: For high-quality surfaces, a change of a tenth of a millimeter often decides the entire impression of an image. That makes them one of the last reliable cases where classic photography stays ahead of AI image generation - not as a matter of principle, but out of physical necessity.
The Pforzheimer Zeitung recently interviewed Thomas Andraschko and me about the relationship between AI images and classic photography. One question from that conversation stuck with me, because it's rarely posed this concretely: for which products does photography clearly stay ahead? The answer comes down to a single word that barely comes up in discussions about image AI: surface.
Why surfaces are their own kind of problem
Polished veneer, brushed steel, engraved sapphire glass – for exactly these materials, the overall impression of an image is decided by changes that play out at the scale of a tenth of a millimeter. Even a minimal shift of the light source changes how a brushed surface feels, whether a polish reads as high-end or cheap, whether a material builds trust or suspicion. That's not a peripheral aesthetic detail – for industries like watches, kitchens, or automotive trim, it's the actual reason people buy: the customer isn't buying the product, they're buying the impression of quality the surface conveys.
That's exactly what makes this category such a tough test case for any image production, classic or generative. Photographers who specialize in these products build up, over decades, a feel for how light behaves on which material – knowledge that rarely exists in words, but in the hand that shifts the reflector two centimeters until the polish "reads right."
Where image AI hits a physical rather than technical limit
The difference between a photo and an AI-generated image lies, as I've described elsewhere, in the process, not the finished result: a photo captures something that's really there, a generated image has to plausibly reconstruct that same physics out of training data. For most subjects, that difference barely shows today. For surfaces with high material demands, it still does – not because the models are fundamentally overwhelmed, but because the requirement for physical precision is especially high here. A slightly wrong reflection on smooth lacquer won't register with a layperson. It registers instantly with a customer who's been buying kitchen fronts for years.
That also explains why the "photography or AI" discussion runs differently for these products than for high-volume catalog goods. There, the technique that scales variants most cheaply almost always wins – I've argued that in detail elsewhere. For materials whose entire value rides on surface perception, that calculation flips: here, photography's higher effort still pays off, because one genuinely convincing image matters more than a hundred interchangeable variants.
What follows for practice
The conclusion isn't "AI out, camera in," but a clearer division of labor than shows up in many discussions. Where a product sells through texture, materiality, and fine light response, the physical shot remains the more reliable path for now. Where environment, variant count, or pure ideation are what matters, generative AI has long had the edge. The skill that counts here is no longer just photographic or technical craft alone, but the ability to recognize, product by product, which case you're actually dealing with – and to decide the image accordingly, not by whichever tool happens to be hyped at the moment.
Conclusion
Surfaces are one of the few areas where photography's edge doesn't rest on nostalgia or taste, but on a plain physical fact: some material properties can still be photographed more reliably than they can be plausibly generated from training data. That will keep shifting as models improve – but as long as a tenth of a millimeter decides the impression of quality, the camera remains the tool that hits that difference most reliably.