>>68285872hm that's a surprisingly tricky question. the TLDR is the model is good. but:
the model here and in the last one is PonyXL which can do a variety of poses and reliably make coherent sex scenes and xrays and stomach bulges and internal creampies etc without the need for a lora or much effort at all desu, which is why i fall back on it for quick throwaway gens or "weird" stuff like splits that i'd have to go find a lora on civitai for normally. despite the name it's currently the flavor of the month model in other threads across the site for genning coomer stuff for normal anime
but it makes very simplistic images with pretty flat shading, even with decent artist loras. usually i try to recover some of the shading and detail that pony doesn't bother with by running a few steps of a second pass with a good old fashioned 1.5 anime model, which usually have better image quality but are way less flexible out of the box, but that can fail even with minimal denoising if the composition is something it just doesn't get. like in this case a standing split
that's why i called that first gen 'lazy' desu, the 1.5 refinement pass was too scuffed to use because i didn't go find a relevant lora to help it along which is why she looks so.. plain. most people probably arent too bothered by it but it sticks out to me pretty badly lol. well i also had a lora i was testing enabled for that pass by mistake still from making
https://files.catbox.moe/njpqo9.jpg yesterday but whatever
anyway i use comfyUI for sdxl stuff due to performance reasons but i find inpainting to be horrendous using it, so as far as post processing goes in that pic in particular the only thing i did was use photoshop to remove an extra finger in one of her hands and make the hairclip transparent instead of solid.
picrel was the original hand, i'd rather draw a little in photoshop than inpaint using comfy or be bothered to switch UIs to inpaint one finger and a hairclip