Just follow the instruction in the workflow
I've tested on Realistic and semi-realistic/anime and it works on them. ComfyUI will download a model for CLIPSeg masking node that detects the face when you first queue a prompt.
sd15FaceUpscale_v10.zip