From a furniture photo to a measured cut list.

You cannot cut a picture

To build furniture you need to know how many parts it has, the size of each one, which board runs past which at every corner, and how many sheets to buy. A generated render answers none of that.

The Fabivo assistant does not draw pixels. It writes a parametric model: a carcass with sides, top and bottom, spaced shelves, dividers, a back, fronts, and a plinth or legs. Fabivo turns that model into boards with exact sizes, and from there it is the normal app workflow.

What the AI does with your photo

The assistant works the way an experienced maker reads a photo:

• Counts rows, columns, shelves and doors, unit by unit.
• Estimates sizes from the proportions in the photo and from common dimensions (bookcase depth, TV stand height).
• Picks a standard board thickness, 18 mm unless the photo shows otherwise.
• Sees its model next to your photo so it can review and correct the shape.
• Can ask about sizes or details the photo does not show. It can still make mistakes: check the result.

Change it by talking or with the mouse

The result is a normal Fabivo model. Ask “make it 20 cm wider” or “add doors to the bottom row”, or drag a shelf in the modeler. You can undo changes. Then generate the cut list again to cut the current version.

Sizes read from a photo are estimates. If you already know the width or height you need, tell the assistant before it starts, or correct it afterwards: the other parts are recalculated.

What it can build

Sheet-goods furniture: bookcases and shelving, TV stands, sideboards, display cabinets, entryway benches, simple kitchen units, desks and tables on legs. Real examples with their cut lists are in the free furniture plans.

It does not do upholstery, curves, metal frames or turned parts. A sloped board, yes; a curved backrest, no.

Measured, not promised

We ran the same assistant with three AI models on 17 photos of real furniture, one attempt per photo. The results, times and costs are published, failures included, in the AI furniture benchmark.