Hello, everyone. Here is an attempt to merge WaifuDivision 1.4 SDv2 model with vanilla SDv2 depth-512 model in the noble pursuit of creating depth aware SDv2 img2img model that actually can depict ?????? as well as other objects. It's the work in progress, but since i don't have much time on my hands, i decided to upload and let it fly like a bird into the sky for you to try.

Usage is simple, you put the image into img2img (or Inpaint for that matter) and put in prompt what you want final image to be, increase steps and it will try to render new image while keeping your subject pose intact. As you can see on examples the concept works pretty good, even with denoise set to 1.

It tends to output paintings, mainly because merge was heavily in favor of anime model that is WD. Also needs to be mentioned that to avoid scary looking faces that all SDv2 models are famous to produce, creative usage of negative prompt is a must. I recommend using Neg_Facelift768 embedding and all kinds of bad things that you can fit to Negative.

Have fun prompting, neuro friends.

TODO: increase quality