TL;DR: it helps to downscale source image first and than upscale it.
Before upscale using IMG2IMG-tool you need for a model to take a "grip" on source image - to add some fraction of model's style to image. At the same time you don't need model to add too much changes to image from model's imagination. So you need to balance denoising strength (DS) carefuly. On low DS important image details stay the same, but too few of a style from model is added.
So how to add more style? Style also added while upscaling, even with low DS - model has to device new pixels from its imagination. So firstly you want to downscale source image to some really small resolution - like 256-384px per bigger side, and then upscale it to about 768px per bigger side with DS about 0.5. And at this point you allready will have pronounced model's style and saved pose and other details from original image. Then you can run several turns on one image "upscaling" it 1 to 1 size with DS about 0.5, adding style and saving pose and other. After that you finaly can upscale it by 3-4 times on DS about 0.35 and there will be good results.
In post below are examples. Them are X-rated. https://civitai.com/posts/2669070
Source image was about 3K per side, I've resized it to 256px, upscaled result to 768px, and from it to 2.3K.