TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 18—21 · POSITIONAL · HELD

Coherence across a set

Somebody has to hold the sky still across the frames, and that somebody is the designer.

The dilemma

Four images for one project, each strong on its own. Together they go on a board. Do you judge them one at a time, in four passes, or do you ask the harder question: do these four belong to one building?

The choices

Approve each image on its merits. Or hold the four together and check three things across the set: is the light from the same sky, is the scale believable from frame to frame, is the same material actually the same material?

The consequence

Four individually good images with two different suns, a door that is 2.1 metres in one frame and 2.6 in the next, and a laterite wall that turns to brick between views do not read as a building. They read as a moodboard. The client senses it before they can say it. Checked as a family, the weak frame shows itself at once. It is usually the one made last, after the brief had drifted.

The case

A student's four-image family for a Coorg homestay. The main view has warm late-afternoon light from the west. The courtyard view has cool overcast light. The detail view has a roof tile profile that appears in neither. Each image was approved on its own. Laid side by side, the set falls apart in ten seconds. The fix is not a better image. It is one line in the brief, "the same 4 p.m. west light in every frame," carried through the whole set.

Our position

Coherence is a judgment made across the whole set, and it is where visual families quietly fall apart. No single image can pass or fail it. The Lab's five curation questions (mood or design, coherent, scale believable, materials consistent, what would need redrawing) are asked of the set, not of the frame.

Why we hold it

A building is one thing. A set of images that claims to show it is claiming to show one thing, and generated images, each drawn fresh from the model's habits, do not keep that promise on their own. Somebody has to hold the sky still across the frames. That somebody is the designer.

The strongest objection

Tools are catching up. Reference-image conditioning, consistent-character and consistent-scene features, and multi-view generation are being built precisely to keep light, scale and material stable across a set. If the machine holds coherence reliably, the designer's check becomes a glance, and this card overstates the work involved.

What would make us revise it

When a mainstream tool produces four-view families that pass the light, scale and material check without correction in most workshop runs, the card is rewritten from "coherence is your job" to "coherence is your check." We test this every edition on the same brief.

Try it

Take a three- or four-image family and check it for consistent light, scale and material across the set. Find the one frame that breaks the family and say exactly what breaks it: the sun, the door height, the material. Then write the one brief line that would have prevented it.

Take it to crit

Ask whether the images in the student's visual family belong to one project or only share a folder. Point at a material and ask them to find it in the next frame.

How it works

Each generation is independent unless you make it otherwise. The model does not remember the sun it used in the previous frame. It draws a plausible sun again. So coherence has to be carried by the brief (the same light, the same material names, the same camera height, stated every time) and by the reading afterwards. A line in the brief asks for the same sun; it does not enforce it. What enforces it is the system's own controls — a reference image the tool conditions on, a fixed seed, a shared camera setting — and your check afterwards. Some tools now let you feed in a reference image to anchor the set. That helps, and it still needs checking.

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.

Age grows from 11 at the centre to 22 at the edge, and six sectors show the learning strands. Tab into the map and the arrow keys step from idea to idea, following the links where there is one. Enter opens the idea under the cursor, and E reads out its links and the reason recorded on each. Press slash for Search, question mark for the full key list, and Escape to leave. Open Ideas for the complete readable list, including what each idea builds on and what it opens up.

LOGIKA · RBDS AI LAB INDIA
ON-RAMP · AGE 11 · FIRST ENCOUNTERS, NOT GATES — IDEAS · — DEPENDENCIES
DONE
OPENS NEXT
SOLID — STANDS ON