>>1751
I've been using img2img recently as well. I'm a lot more of an armature when it comes to 3d modeling / koikatsu but I've had some decent results. This one isn't ryona or guro but it shows my results. I haven't tried guro img2img yet but it should work in the same way.

A bit tip I can give is to use inpainting with "only masked" instead of "whole picture". Regardless of the image size I set it to "only masked" and the resolution to 1024x1024. It will render out the masked section in 1024x1024 and then either downscale or upscale it back to fit exactly where it should be placed. Obviously you want it to downscale it, but it will work even if it is somehow upscaling it.

To give a rough run down of my workflow for img2img. I first do an img2img pass of the whole image with a low denoise, I use around 0.2~0.4 since ponyXL is aggressive. This is for the background, if I'm getting too many errors I place a mask excluding the characters. For the mask I re-render the same image with no background then use magic wand on photoshop to pickout the background to make the mask. I don't how this would work on ComfyUI but automatic1111 has mask uploading. I then go over each part of the character and inpaint them using "only masked". This can be done with either a low or high denoise. I typically section them off by face-body-any "complex behaviour". The danger of using "only masked" is that it can't understand the context of the whole image so skin tone for example may differ if you do the body in multiple "only masked" inpaints. If you need context I like to do a "whole picture" inpaint generation then go back to "only masked" with a low denoise just to fix up the details.