ControlNet is a family of techniques layered on top of diffusion models that lets the user constrain what stays the same. The idea: instead of giving the model only a prompt, you also give it a structural map — Canny edges, a depth map, a pose skeleton — and the model is required to match it.
For interior design, the relevant controls are edge detection and depth. Edge ControlNet preserves window placement, door frames, and architectural lines, so the redesigned room sits inside the same shell as the original. Depth ControlNet preserves the sense of space — what's near, what's far — so the new furniture feels at the right scale.
Without ControlNet (or a similar conditioning technique), img2img drifts. The model might turn a bedroom doorway into a bay window, or compress a long room into a square. ControlNet keeps the bones of the room in place while letting everything inside it change.