V2 UPDATE: Works much better with anime, various improvements to the mix!
UPDATE: Made a LOCON that can given even better results with this checkpoint https://civitai.com/models/139088?modelVersionId=153856
Working in latest commit of AUTOMATIC (1.6):
For best results use refiner 1.0 no VAE which you can find here: https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0/tree/main
I use a "fixed" VAE that improves speed and reduces errors on FP16 here:( https://civitai.com/models/140686?modelVersionId=155933 (sdxl_vae.safetensors)
[SDXL-VAE-FP16-Fix is the SDXL VAE*, but modified to run in fp16 precision without generating NaNs.}
This mixed checkpoint gives a great base for many types of images and I hope you have fun with it; it can do "realism" but has a little spice of digital - as I like mine to.
https://www.patreon.com/nucleardiffusion
https://www.redbubble.com/people/frodosart/shop
Accepting Commisions!
https://ko-fi.com/nucleardiffusion