A Room Has to Be Walked Through, Not Only Looked At

How it works

The photograph is scaled to about one and a fifth megapixels and repainted at a fairly high denoise, because furniture is being moved and replaced rather than restyled.

The instruction is unusual in this catalog for being about use rather than about appearance: every doorway stays clear and reachable, there is walking space between the furniture, seating faces the room's focal point, and each piece sits at a usable distance from the others. The architecture and the camera viewpoint are held.

The negative prompt is a list of things that make a room unusable rather than ugly — furniture blocking a doorway, no walking space, seating facing a wall, furniture floating in the middle of the room.

Design is constraint, and a model has none

The thing a designer is actually paid for is not taste. It is working within a list of facts: the budget, who lives here, what the room has to do in the evening and at eight in the morning, where the sockets are, that the radiator cannot move, that one occupant is left-handed and the other watches television.

A generative model has none of those facts, and cannot ask. What it optimizes is whether the picture reads as a well-composed interior, which is a different objective that coincides with a good design often enough to be misleading.

This is why generated interiors converge on the same look. Asked for something good with no constraints, a model returns the center of what it has learned, and the center of interior photography is a narrow and well-lit place.

The productive way to use it, then, is as something to react against. A proposal you dislike tells you a constraint you had not articulated — that the room needs to seat five, that the wall has to stay clear for the door, that you would never sit with your back to the window. Those sentences are the design work, and the render is what provoked them.

The checks are numbers

Unlike style, circulation is measurable, and the render can be tested against numbers rather than opinions.

A main route through a room wants seventy to ninety centimeters. Less than that and two people cannot pass, and one person carrying anything cannot pass at all.

A coffee table sits forty to fifty from the sofa — close enough to reach, far enough to get past.

A dining chair needs about seventy-five behind it to be pulled out and sat in, which is the clearance most often missing in a render, because a pushed-in chair looks tidy.

A door needs its full swing, and an inward-opening door eats a quarter circle of floor that no picture makes obvious.

Take one known dimension from the photograph — a door is about seventy-six centimeters wide — and use it to scale everything else. Then check the four numbers above. That is a five-minute test and it is the difference between a mood and a plan.

Where it sits with the other room pages

Five pages here change a room and they answer different questions, which is worth having straight before running any of them.

Wall repaint changes one property and keeps the light. Furniture replacement changes one object in a room that otherwise stays yours. Empty room staging furnishes a bare room, with scale as the thing that must not lie. Room redesign changes the style of everything inside the shell.

This one changes where things are. It is the only one of the five whose output is judged by walking rather than by looking, and the only one where the failure is invisible in the picture by construction.

Publication gate

This page ships once the workflow has been run against a living room with two doorways, a narrow room shot from the door, an open-plan space, and a room with a fixed radiator and a bay window, with a door width used to scale each render and the four clearance figures checked in every case.