Try size: 640 x 640 (SD1.5) / 1024 x 1024 or 768 x1024 (Pony and ILXL and OOO)
Weights: 0.8~1
Clip skip: 2
Training material from bing
Pictures from Calicomix 7.5 and AnimeXL-xuebiMIX and CalicomixEroill_v30 and OOO_v80
Note: If need cover, please delete "(((red R18 word logo)))," and add "(((portrait photo cover))),", then use "Hires. fix". Fingers and R18 are not very stable. (SD 1.5)
Note: If need dress and guest, please delete "blue cap, blue polo shirt, pants,", then add "crowd". Since 4/17, I have changed the file suffix name from "_sdxl" to "_pony". (Pony)
Note: I adjusted the "Rate of caption dropout" from 0 to 0.2, trying to get Lora to add some random variations instead of generating similar images every time. Because when we overtrain the danbooru tag, it will cause a "canned prompt effect". If ILXL's sensitivity to prompts is reduced, it will be more expectant like Pony. (ILXL Fix)