Post-AI Imagery Fix

November 16, 2025

a theory of visualization, image as architecture, image implant

I’ve always been interested in how the idea of a building is communicated visually to clients and have written about it at least twice – the first time in Keeping It Real. I’m not that interested in how buildings are communicated visually to the media because that’s career building as opposed to building building. However, the careers of some architects are driven by the production of images for those of us who consume them. The images we see are most likely not the ones seen by the client’s project manager. Selling a project to media is different from selling it to a client but selling a project to a client might sometimes depend on whether it will sell to the media and, in turn, the public. The word for this is reflexivity, which refers to circular relationships between cause and effect, especially as embedded in human belief structures and Architecture which, despite Schumacher’s protestations, is one, [c.f. The Mystery of Beauty] and Architectural “discourse” its worship.

Even this next drawing produced for the US embassy in Berlin by Moore Ruble Young in 1995 has an air of agreed unreality about it. It’s an example of what used to be called an “artist’s impression” because that’s what they were. The watercolor below is most likely an image for public consumption a.k.a. perception management, especially because it’s a high-profile building in a prominent location.

In 1995, computer rendering was still in its infancy. ArchiCAD was the only architectural application with native 3D visualization capability but, by 2005, most packages offered some degree of it. Those capabilities were redundant by the early 2000s when architects were studying massing in Sketchup and presentation renders would be produced by “the render guy” who was adept at the increasing powerful rending engines available. Designers would export their files to the render guy who would set viewpoints, textures and light sources and levels. Artlantis was the last such software I ever used but other engines such as AI Visualizer, Cinerender/Redshift, Twinmotion, D5 Render and Cinema 4D were newer CPU-powered alternatives. Enscape was the last of these I remember. Horrible things were produced in the name of photorealism, much of it involving cherry blossom trees and children with balloons but, in general, photorealistic was the ideal, and ways to produce more photorealistic images were duly invented and marketed. Dark times.

About 2016 there was a limited backlash with post-digital rendering as announced on Metropolis in 2017 and ArchDaily in 2018 and again in 2021 by which time we were familiar with its use of Photoshop collage and fanciful textures and patterns applied to a photograph or a photo-real base image. You know the look.

Post-digital soon became bogged down with references to Hockney, Magritte and de Chirico for artistic cred, and then with Superstudio and Archizoom for architectural cred. Still, it was a way to make images that, in the eyes of their creators, said something about concept, atmosphere and other intangibles to clients and public alike when, only a short while back, we were being told clients wanted more photorealism. Post-digital collage remains a good way for students to produce atmospheric images without too much time and effort but, as with anything, there’s the style vs. content trap. Post-digital representation doesn’t mean rolling the clock back to a time before photorealistic renders but situating itself on the other side of digital representation in the sense of restoring some quality that we didn’t know we’d lost. Finding something “refreshing” is usually a sign of this. Post-digital then, isn’t nostalgia but a kind of backtracking to the last familiar place before choosing a new path ahead. I like to think of this as the lost-in-the-woods analogy. You follow your markers back to the place where you took the wrong fork in the road. It’s like having to go backwards in order to go forward again. If only we could do this with architecture. Instead of wondering which way to go next we might better off looking for where we took a wrong turn. Let’s do this some other time. Meanwhile, client tastes were to change once again when animated images became the new way to keep one’s processor warm. “Animated vegetation impresses clients and conveys solutions to audiences who might have trouble fully conceptualizing your ideas.” Let’s not dwell on this.

The second time I wrote about renders was with respect to the emotional layer. [c.f.. The Emotional Layer] With great computing power comes great responsibility and, sure enough, somebody noticed that, for all their pixels, digital renderings were somehow dull and flat and that the solution to this was to add an emotional layer. This next is an example. You’re too busy looking at the forlorn bird to even notice that the topography, trees and rocks are somebody’s imaginings. I understand that architects need to communicate their design intentions to clients and other parties and maybe that’s what’s wrong with this image. The house is already and, were there people living in it, they’d be inside and warm when it was cold outside. This image has no purpose other than to say somebody made it.

Having more emotion in architectural renderings quickly wore off once AI could generate images that could, at a stretch, become buildings. AI isn’t known for its emotional sensitivity and I suspect this has something to do with why “the emotional layer” isn’t something we’re told to aspire to anymore. I suspect that, even as I write this some coder is pondering how to code to make faux-emotion appear more authentic yet specifiable at the same time. Meanwhile, we’re being told that clients now like lots of options and how fortunate we have this thing called AI to generate them. Here are three options. They’re identical. Nobody cares which one you prefer.

Weeds are said to be just plants in the wrong place. Ugly buildings might just be buildings in the wrong place but, with these fudged images, it’s hard to tell. Modernistic interventions contrasted against nature is so 1930s. The only association wth context here aligns with what Zaha Hadid [R.I.P.] said about curves somehow being “natural” and what artificial things should aspire to be.

At first I was skeptical about AI in architecture but now I’ve become more wary of talk about how it’s just a tool to enhance creativity while increasing productivity, whatever that means. I’m beginning to wonder if the real problem isn’t how awful AI architecture is but how awful AI-enhanced creativity is. When I look at images like these I’m seeing decisions made by humans. These aren’t AI hallucinations despite their fever-dream vibe. Some human intelligently prompted the process to produce these images and then intelligently stopped the process when they thought the results suitable (for whatever purpose). The AI was just doing what it was being instructed to do and it seems that that was no more than producing images representing somebody’s idea of what AI can do.

So what do we do? Where do we go for our fix of architectural imagery now AI’s not delivering on the emotional and engagement layers? Right now I’m charmed by images of buildings in manga style because I can enjoy images of buildings and their contexts as well as a story to which the building may well be refreshingly incidental. It’s all I need if I’m never going to visit or use the actual buildings. All that’s needed to create this architectural content are imagination, ink and paper – plus the energy provided by pizza and caffeinated drinks. There was an exhibition, MANGARCHITECTURE: Absence of an Architect at TOTO Gallery Ma [link]. Yasutaka Yoshimura was the architect and he was absent because he suffered a brain aneurysm [yikes!] just prior so he asked some manga artists to incorporate his projects into manga that became the focus of the exhibition. And a book. Away from the slop, I’ve found my happy place.