Generative systems can produce a convincing building in seconds: surfaces, light, furniture and dramatic viewpoints. Yet architecture is not exhausted by its image. In Wim Wenders’s immersive film about Peter Zumthor, space emerges from matter, time, bodies and memory—the elements most resistant to instant generation.
Generated architectural imagery privileges the final moment: a perfectly lit building without construction, maintenance or conflict. Zumthor works with the duration before and after that image. Materials chosen to age, details transformed by light, surfaces that retain traces and paths that slow the body make design temporal. This continuity, difficult to reduce to one frame, separates an imagined space from a lived one.
Zumthor’s work is experienced through weight, temperature, acoustics, smell, texture and the time required to cross a threshold. Wenders’s 3D camera does not merely document façades. It tries to place the spectator inside relationships of distance and presence. The film’s technology serves slowness rather than spectacle.
Wenders’s stereoscopic cinema approaches the problem from another direction. Three-dimensionality is not only spectacle; it conveys distance, volume and presence. Viewers cannot touch the material, yet they sense the proximity of a wall, the movement required by a threshold and the way light crosses a room. The camera becomes a substitute body, measuring the relationship between architecture and observer rather than simply describing a façade.
AI images often reverse that hierarchy. They begin from appearance and infer a plausible building behind it. The result may be seductive while remaining spatially incoherent or materially impossible. A staircase can look correct without leading anywhere; stone can behave like plastic; a room can produce an atmosphere without offering a body a place to stand.
AI can still be fertile for architects when placed at the right stage. It can multiply hypotheses, test atmospheres and challenge a first solution. Trouble begins when persuasive visualisation is mistaken for verification. Structure, access, climate, acoustics and material behaviour cannot be inferred from beauty. Generation should remain a question addressed to the project, not proof that the project works.
This does not make AI useless for architecture. It can expand sketching, compare alternatives and help communicate an intention before construction. But an image is not yet a space, and optimisation is not habitation. Architecture begins where resistance appears: gravity, weather, regulations, labour, ageing and the unpredictable ways people occupy a place.
There is responsibility toward construction as well. Instant images easily erase labour, craft, material origins and site conditions. Real architecture emerges from negotiations, errors and adaptations no render reveals. If AI enters the process, it should help make these relationships visible rather than produce seductive surfaces alone. Innovation does not mean removing matter; it means understanding matter more clearly before transforming it.
The deepest lesson concerns the limit of vision. We can know a building through photographs, films and models, but no representation equals inhabiting it. The body brings scale, effort, orientation, memory and desire; it changes space by crossing it. AI can simulate observers and predict routes, but it does not yet possess a physical biography with which to remember a place. Architecture and cinema retain a profoundly human function in that distance.
Wenders and Zumthor offer a useful limit for generative culture. The machine can imagine endless visible forms, but it does not inhabit them. The task of architects and filmmakers is not to reject synthetic images, but to return them to matter, duration and human scale. A place becomes architecture only when someone can enter it—and remember having been there.