There are two ways to make a picture of a nicely styled room. One is to generate a new room from a text description. That is easy and it is not what anyone wants, because it is not their room. The other is to take the photo they gave you and change what is in it while leaving the architecture alone. That is what we do, and it is harder for good reasons.
Read first, render second
Before anything is rendered, a vision model reads the photo and writes down what the room is: its shape, where the windows and doors are, what light it gets, and which pieces are worth keeping. That read is not shown to you, but it is the spine of the redesign. The render prompt is built from it, and it leads with the room, not the style.
Then the image model edits the photo, at high input fidelity, so the geometry of the space survives. The instruction is blunt: do not move the walls, the windows, the floor, or the camera. Re-dress everything else. The result is your room from your angle, in a new style.
Why a separate piece list
A picture is a direction, not a shopping trip. So every redesign comes with a list of piece categories that would get you the look, grounded in the room the model just read. Not brands, not a checkout, just the kind of thing to look for. It keeps the render honest: this is achievable, here is roughly how.