A bathtub in a desert. An elevator shaft with its concrete walls turned to glass. A telescope the size of a building, assembling itself against a black field. A Brabus G-Class on a mountain road. Forty-three seconds, cut in July, and on the face of it a collection of objects with nothing whatsoever in common.
They have one thing in common, and it is the entire argument for the discipline. None of them can be photographed.
What product visualization is actually for
The telescope does not exist yet. The Giant Magellan and the Thirty Meter Telescope are instruments still being built, and the only way to show how their mirrors come together is to build them in 3D and film that.
The elevator exists, but nobody can see it. A machine-room-less system spends its working life inside a concrete shaft, which is why that sequence replaces the shaft wall with a transparent proxy. The engineering is the product, and the product is invisible by design.
The bathtub had no prototype when the images were made. Nor did the finishes on several of the sanitaryware ranges in the reel, which is the ordinary condition of a launch campaign: the catalogue closes before the first unit leaves the line. As Botond puts it, “we used to think all the photos in IKEA catalogs were real. Turns out, they weren’t.”
And the car could be photographed, in principle, on a road that is closed for a day in weather that cooperates. That is the version of this argument people underrate. The question is rarely whether a camera could. It is what the shoot costs when a brand needs eleven colourways, three continents and a reshoot in six months.
The uncomfortable number, with its caveat intact
Industry research across more than a hundred studios puts CGI at roughly six times more cost-effective than photography over a typical launch cycle, and the saving compounds with every variant, region and refresh.
The caveat matters more than the number. For a single shoot of a finished product with no variants, photography is often still cheaper and faster. Anyone who tells a brand manager otherwise is selling something. It is precisely because that exception is real that the rest of the case holds up.
Meanwhile the demand side has moved. The number of images a buyer wants to see before purchasing online has risen from three to eight per product. Eight images per SKU, across a catalogue, refreshed each season, is not a photography problem. It is a production-line problem.
Where the AI line actually falls
ZOA uses AI in architectural work, openly, and argues it is a genuine gain at concept stage. In product work the line is drawn much more precisely, and the precision is the point.
AI is used on the environment, on the entourage, and in the animation. It is never used on the object itself, and never on the initial render. That base render has to be super sharp and spotless, because it is the thing a brand signs off as true.
Botond Sass, who heads the product visualization branch, puts the reasoning plainly. “AI adds atmospheric noise or detail that doesn’t exist,” he says. “In other fields, that can be of great use. But for products, it’s pretty much a dealbreaker.”
Read that precisely. Every pixel of the object has to correspond to a real thing. A hallucinated weave in a fabric, a screw head of the wrong type, a logo that is almost the brand’s, and the image is no longer evidence of anything. A product render is a promise about a thing a customer will eventually hold. A generated sky forty metres behind it makes no promise at all.
This is the sharpest distinction in ZOA’s work, and it is a working method rather than a prohibition: AI on the world, never on the object, base render spotless. Very few studios will say where their line falls with that much precision, and a studio that claims to use no AI at all is either not telling the truth or not using the tools.
Control, and the test that exposes it
“At its core, product visualization is about control,” Botond says. “It’s about crafting every surface, highlight, and reflection until the image becomes indistinguishable from a photo – often even cleaner and more compelling than reality itself.”
There is a standard test brands use when hiring a team, and it is unforgiving: hand over a studio-lit photograph and ask for it to be rebuilt entirely in 3D. “That’s their way of seeing how well someone understands realism,” he says. “Getting it just right – the fabric texture, the light falloff, the camera depth – isn’t easy. But it’s what sets great work apart from service level.”
Architecture is more forgiving than this. A render of a building has weather, trees, a city, a sky, a dozen places to put atmosphere when a detail is not resolved. Botond’s argument is that having none of that is an advantage. “Product rendering offers few visual tricks to fall back on,” he says. “These limitations push us to be more creative – especially with composition and post-production.”
The elevator film makes the standard concrete. On manufacturing-grade geometry, if a bolt lands two millimetres off its target, it shows.
Why this is the line that grew
Architectural practices absorbed an enormous amount of image production in-house over the last two years, and that work narrowed to the moments where being wrong is expensive. Over the same period the product line expanded: Villeroy & Boch, Schindler, Brabus, two observatories, marine, furniture, flooring.
It is the same studio and, more to the point, the same skill. Twenty years of material literacy in close-up, pointed at objects half a metre from the lens instead of buildings half a kilometre away. The speaker did not change. The subject did.