From my guess is your pre processor should be none and then you put in the depth map block image as the guide image and set your controlnet model to this model. Just pure guessing, gonna try it when im at home tomorrow

@spoffninja Yes it's what I do... No preprocessor, loosecontrol model set as model in controlnet. the depth image for guidance in controlnet image. I tested with various cfg scale and controlnet weight. Thank.

Blender is lovely for it.

Install Blender. Run it. Render the image in the default document that starts with Blender.

@Vendaciousness Meh, I don't know, if I'm going to boot up blender, I might as well grab actual 3d models of objects.

@UndefinedUsername Well, to be fair, you only asked how to make a box. And I got all excited, because I was finally able to help someone else with a Blender question. Usually it's the other way around, as nothing is intuitive, so I could probably sit and bang my head against a problem for 3 years and not figure it out. But the price is right. ;)

It loosely follows depth maps. While examples are shown with simple maps generated from blender, i'm sure you could use the usual depth estimation preprocessors and it will loosely follow those as well.

that's my theory at least.