TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 15—17 · POSITIONAL · HELD
A green-looking render is not a green building
Leaves on a façade are a pattern, and heat gain through that façade is physics.
Our position
Image models learnt what "sustainable" looks like from photographs: timber, planting, louvres, a green roof, soft light. Whether they learnt anything about what those things do is contested; what is certain is that a picture of a louvre is not a measurement of the heat behind it. Visual plausibility is not performance verification. A render can carry every sign of a green building and still describe a glass box that cooks in April. Leaves on a façade are a pattern. Heat gain through that façade is physics. The machine never connected the two. A jury that takes the leaves as proof of low heat gain has been fooled by a look.
> The render shows what green buildings look like in pictures. It cannot show what this one will do in May.
Why we hold it
A diffusion model makes images that look right. It does not model behaviour: it has no sun, no thermal mass, no wind. The photographs it learnt from were chosen to present sustainability well, not for measured performance, so the pattern it holds is the marketing pattern, not the building's. The general rule is the seductive-image test on this map; this card is that test applied to green-looking images.
The strongest objection
Patterns are not empty. The features a model reproduces — deep overhangs, planting, shaded openings — are in the photographs because they worked, often enough, in real buildings. An image that shows them may steer a student toward sound moves before any simulation runs, and a tutor who dismisses the render as "only a pattern" may be throwing away the only environmental thinking the student has done so far.
What would make us revise it
An image tool wired to a physics engine — where the façade in the picture has been sized by a solar or thermal calculation the student can open — would close the gap this card names. When a mainstream tool ships a render with a checkable heat-gain figure attached, we rewrite this card from "pattern is not physics" to "read the physics behind the pattern." We check what the tools ship every edition.
Try it
Generate "a sustainable house in a hot climate." List every green-looking feature in the image. Next to each, write what it would actually do at your site — shade, cool, ventilate, nothing — and whether you can tell from the picture. Most rows will end in "cannot tell."
Take it to crit
Point at one green-looking element in the student's render and ask what it does at 3 p.m. in April, and how they know. If the answer describes the picture, ask again.
How it works
This card is the environmental version of two cards on this strand. The seductive-image test (`dt_seductive-image-test`) asks of any image, "what decision does this let me make?" Pattern-is-not-precedent (`dt_pattern-vs-precedent`) separates a studied building from an average of many buildings. Here the average is of how green buildings are photographed. A diffusion model is trained on image–caption pairs. Captions with "sustainable" or "eco" sit beside photographs that share certain features, and the model learns those features. Whether the building in the photograph performed was never in the caption, so it was never in the training. The physics — solar gain through glass, mass that stores heat, openings that let wind through — lives in weather files, simulation and measurement. That is where the next cards take you.
What this idea builds on
What this idea opens up
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.