I tried Stable Diffusion XL and these are my thoughts (Patreon)

The image you see here was generated using Blue Pencil XL. The only thing I had to do is to inpaint the face which was affected by the original face ratio of the girl I generated. Because my koikoi model is based in version 1.5 so it wouldn't work well with SDXL.

I will consider training my model for SDXL, but using it for anime is still pretty far. I'm planning to try NovelAI Anime v3 on their official website. But first I'll look for videos of people working on it to see how much freedom I have with it. It's very likely that I'll try some mixed works with SDXL and SD 1.5.

These thumbs up was a test I was doing and most models could handle it pretty well. Around 75% of the time the hands were correct. Also, another thing that I've noticed, the speed which SDXL does img2img is likely to be 5-10x faster. So this can be something that I can adapt to my workflow.

The only problem is that SDXL is waaaay better for 3d, semi-realistic and realistic images. So that's why I want to see the new NAI model myself.