Why do curtains in AI room renders hang wrong?

Curtains are unusual among interior decisions in that half of the decision is a colour and half of it is arithmetic. AI handles the colour half well and is absent from the other half entirely, and because a render looks finished, people do not notice which half they are being sold.

The four things that go wrong

• The drop. There are four legitimate hem heights: at the sill, on the apron 10 to 15 cm below the sill, clearing the floor by a centimetre or two, or long enough to break or puddle. A render has no reason to prefer any of them, so it lands wherever the picture looks balanced. A hem floating five centimetres off the floor is not a style, it is the model failing to commit.
• The fullness. Curtains read as curtains because there is roughly twice as much fabric as the space they cover, conventionally 2 to 2.5 times the track width. Models paint folds onto a flat surface rather than gathering real cloth, so renders drift flat, and a flat curtain looks cheap regardless of the fabric.
• The stack-back. Open curtains have to go somewhere, and a pair stacks into roughly a quarter to a third of the track width in total, more in a heavy interlined fabric. Renders show panels pinned neatly to the edge of the reveal taking no space at all, which is why people end up with curtains that permanently cover a slice of the glass.
• The heading and hardware. Pinch pleat, pencil pleat, wave, eyelet and tab top are different products on different systems. Renders blend them, and the top 20 cm of any curtain image is where the model has the most freedom and the least training signal.

How to still get value from it

• Use the render for what it is accurate about: colour, fabric weight and opacity, and how a length of that colour behaves against your own wall in your own light. A 15 cm shop sample cannot answer any of those.
• Use it to settle hanging height, which is a pure proportion judgement. Generate the track at the ceiling and just above the reveal, and the argument ends immediately.
• Ignore it completely on measurements. Take track width, drop, and the wall available each side, and hand those to whoever is making the curtains.
• Check for a radiator under the window. It usually rules out a floor-length drop, because heat then runs up behind the fabric and out through the glass. No render has ever noticed a radiator.

One useful test before you trust any curtain render: look at whether the window itself has changed size or the wall above it has grown. That is the model reverting to the generously proportioned window it saw in training rather than the one in your flat, and if it has moved the window, nothing else in that image is about your room either. The full version, with the four drops, the stack-back arithmetic and the radiator problem drawn out, is in our guide to AI interior design for curtains.