A simple version trained from some leaked official resources.
Facial marks and accessories still can't be produced consistently.
v2.0 update: it can now produce the earring and facial marks.
tacet mark v0.1 update: used to generate tacet mark on the tongue, trained for inpaint purposes, but sometimes works on regular txt2img too.
v2.1 update: improve stability when using other models.
Generate 1024 x 1024 images directly and use Ultimate SD upscale to scale the image to 2x and then 1.5x again.
Facial marks and earrings do not require inpainting. (Sketching and inpainting still can be used for improved accuracy.
I've written a simple tool to help clean up the prompts and I hope it helps.