Method

This page says what DecoScale measures, how, and what each number is worth. Every figure below comes from measurements made on real rooms surveyed with a tape measure, or on synthetic scenes whose geometry is known. What has not been measured is not written.

Three photos, taken from three different spots

A photo shows what it frames, and nothing more. A view taken from a corner of the room cuts off the length of the facing wall: the measurement that comes out is that of what is visible, not that of the room. Three photos taken from three different spots complement each other, each seeing in full what the others cut off, and for each dimension the measurement keeps the view that frames it completely.

What matters is to move, not to turn on the spot. A panning shot adds width, never depth. We checked it by filming five sweeps of rooms surveyed with a tape measure:

panning video	17 to 48 % error on the dimensions
three photos, three spots	1.7 to 2.8 % on the same rooms

Measured on 25 August 2026 on a kitchen and a bedroom. Neither resolution nor format changes anything: it is the framing.

What a photo lets you measure, and what it does not

Asking an artificial-intelligence model “how wide is this room?” does not work, and we measured it before writing it: on synthetic scenes where the truth is known, the median error is 219 %. The model reports the width of its own field of view, not that of the room. Depth and ceiling height, on the other hand, read well.

So DecoScale does not ask that question. It rebuilds the room in three dimensions from each photo, with a metric depth model, then derives the floor from it. On a view that frames the whole room, the median error is 6 %; merging three views on closed rooms surveyed with a tape measure, it is 12 %.

The limit of this measurement is honest and worth knowing: the lens measures what it sees. A door open onto a hallway, a room open onto the entrance, a bay window: everything visible beyond the walls enters the measurement and enlarges it. That is why the screen lets you draw, on the photo, where your room ends.

Two tape-measure dimensions, not one

The automatic measurement is a starting point, not an answer: the screen shows you what it found and offers you two tape-measure dimensions, the width of the back wall and the depth. Two, because one is not enough, and that is the point most tools keep quiet about. You can also continue automatically: if the reliability of the measurement is below 0.7, DecoScale first warns you that some furniture might not fit in its place, and you confirm.

A single dimension corrects the scale of the room, but not its shape: the ratio between length and width stays the one that was estimated. Measured on six rooms with a tape measure, a single dimension leaves a median error of 19 %, and up to 59 % on a hallway. With both, each axis comes from you, and the room is exact.

The margin is shown, never promised

You will not find on this site any sentence like “±5 % accuracy”. That is not caution, it is a measurement: we found that a model’s self-assessment never covers its real error, and that a fixed margin is wrong in both directions, too pessimistic on a simple bedroom, too optimistic on a hallway.

The margin shown is therefore computed for your room, from the spread between your three views. And because an estimate drawn from a photo never deserves full confidence, DecoScale does not act as if the room were known: it warns you before continuing when the reliability is below 0.7, and offers you the two dimensions to make it exact.

Real furniture, at the dimensions the merchant publishes

Every object in the render is a retail product, with the dimensions and price the merchant publishes on its page. Before being suggested, it has to fit: in the room, but also in front of a door, under a window, away from a radiator, with room to walk. These are not rules of taste, they come from a survey of interior views.

When a product is only a close match to what the image shows, the render says so. When a price is “from”, the render says so too. With SLF24 and Massivmoebel24 the links are affiliate links and we earn a commission paid by the merchant; elsewhere the link is direct. Either way the price is the merchant's, and the choice of furniture depends on nothing but your room, your budget and your style. The commission enters no selection calculation.

The render is an image, and it is checked

The final image is generated from your room emptied of its furniture, with the selected products placed at the computed positions. An image model never flags its own mistakes: it always renders a plausible scene, even if it has erased a door or invented a window. A computed check therefore compares the render with your photo, fixed element by fixed element, and when one of them has vanished, the render is redone, provided the redo is strictly better.

The instruction given to the render is to keep the shape of your room: no wall moved, no opening added. It is not perfect, and we measure it: on our benches, roughly one render in eight still adds an element that does not exist, and every change of instruction is measured over eight draws before going into production, because a sentence that sounds cautious can make things worse. A style whose instruction cannot hold in your room, for instance a loft with a glass roof in a bedroom without a window, is not offered to you.

How long

Expect one to four minutes between uploading the photos and the image, depending on the room. The image appears as soon as it is ready; the clickable areas arrive between thirty seconds and two minutes later. It is not instant because every step is real: analysing the photo, measuring in three dimensions, clearing the room, selecting from the catalogue, laying out, rendering, then locating each product in the image.

What it is worth, in one sentence

DecoScale does not know how to measure everything, and it does not pretend to. It measures what can be measured, asks you for what cannot, shows the uncertainty that remains, and only suggests furniture that exists, at the dimensions its seller publishes. That is the whole method.

On how your photos are used, see the privacy policy. To get started: create a layout, the first one is free.

Frequently asked questions

Why do you need three photos taken from three different spots?

A view taken from a corner of the room cuts off the length of the facing wall. Three photos taken from three different spots complement each other, and for each dimension the measurement keeps the view that frames it completely. What matters is to move, not to turn on the spot: a panning shot adds width, never depth.

Why two tape-measure dimensions, and not just one?

A single dimension corrects the scale of the room, but not its shape. Measured on six rooms with a tape measure, a single dimension leaves a median error of 19 %, and up to 59 % on a hallway. With both, the width of the back wall and the depth, each axis comes from you, and the room is exact.

What is the measurement from the photos worth?

On a view that frames the whole room, the median error is 6 %; merging three views on closed rooms surveyed with a tape measure, it is 12 %. The lens measures what it sees: an open door or a bay window enlarges the measurement. The margin shown is computed for your room, from the spread between your three views.

Can the render get things wrong?

Yes. An image model always renders a plausible scene, even if it has erased a door or invented a window. A computed check compares the render with your photo, fixed element by fixed element, and when one of them has vanished, the render is redone. On our benches, roughly one render in eight still adds an element that does not exist.

How long does it take to get the render?

One to four minutes between uploading the photos and the image, depending on the room. The image appears as soon as it is ready; the clickable areas arrive between thirty seconds and two minutes later.