>>5154

>depth doesn't work great on images that are too flat

I've had a few such cases, too. Especially when working off of uncolorized lineart. What helped me greatly was to just quickly and crudely add some simple single tone shading.

If you are looking for an easily useable freeware program for that sort of photoshopping stuff I can recommend paint.net (not a website, the name is just a bit confusing). With that I'd just load the sketch picture, add a layer with its mix mode set to "multiply", so that the lineart doesn't get deleted when you brush the shading over it. The shading really doesn't need to be precise, Anytest can chew it up even if it's all over the place and out of the lineart bounds.

Something else you might want to try for stuff like the shotgun/rifle to her head in your last pic; you can look for a fitting 3D model of a weapon on sites like sketchfab. You can rotate, shift and zoom the models in the preview window there, take a screenshot, remove the background and copy that into your template image. If webui's built in background removal has problems with that you may have to reduce the noise and possibly adjust the brightness/contrast first (can also both be done with paint.net). Once the background is removed you can just drop it into your template image (best use a new layer). Now depending on the controlnet weight settings the weapon (or whatever object you may add this way) might get scrambled into some AI-weirdness, of course. But in most generations the rough outline will be filled with something. But if you get an otherwise good image you can easily fix that with inpainting in img2img. Activate Anytest in controlnet again, check "upload independend control image" and drop your original template image with the photoshopped weapon/object in and set the controlnet weight to something well above 1 (you may have to experiment to find a good setting). once you got a good resembling image you can run that again with lowered settings to get it to match the style of the rest of the image. Of course if you don't need the object to be present in the initial txt2img step it becomes easier and you can just drop the object into your pic and directly use that with anytest in img2img. Set the denoise to around 0.4 to 0.5 and the controlnet weight to ~0.8 - ish.

Case in point; <=== hope you don't mind I gave her an AK47. ^^ ..The original looked like one of these things: https://www.lematec-tools.com/images/148794 ...that would give her a good headache but unless you'd also push her head against a wall with it and kept pushing really really hard you might be disappointed with the results. xD

>currently I'm using ntr illustrious v4 model with sd webui

Have you tried Forge instead of webui, yet?