How AI interprets · 4 min read
How AI interprets brutalist architecture.
On the consistent failures of AI-rendered brutalist work — and which Simon Phipps photograph it tries to imitate every time.
Image models trained on broad corpora produce brutalist architecture that reads as the genre's greatest-hits Pinterest board — same Trellick Tower angles, same Barbican Estate light. The default is recognizable rather than studied. Directed work fixes this by specifying surface texture (board-formed, exposed aggregate, shutter marks) and light angle, not just the word "brutalist."
Brutalist architecture is having a moment in image-model output. The result is recognizable but unstudied — endless Trellick Towers, endless Barbican walkways.
What models default to
- Trellick Tower three-quarter view (Erno Goldfinger, London, 1972)
- Barbican Estate elevated walkway (Chamberlin, Powell & Bon, 1976)
- Habitat 67 modular cluster (Moshe Safdie, Montreal, 1967)
- Boston City Hall plinth (Kallmann McKinnell & Knowles, 1969)
What gets missed
Surface specificity. Board-formed concrete shows the grain of the wood that shuttered it. Exposed aggregate shows the gravel. Shutter marks show the seam pattern of the form. These are what distinguish photographed brutalism from rendered brutalism. Models default to a smooth grey wall.
How Quiet Geometries fixes this
Each plate specifies surface texture in the brief. Plates 01 (Concrete stair) and 07 (Corner) require board-formed surfaces. Plate 03 (Water tower) requires the shutter pattern of post-war British water-tower construction. Plates without the surface specification do not survive.