AVAILABLE FOR LEARNING ONLY.

NO COMMERCIAL USE, NO REUPLOAD, NO R18G USE.

Trained on 200 illustrations by Xin&Obiwan from Danbooru.

Using device on AutoDL (EPYC 9654 + RTX4090 ×6).

GPU=6, Repeat=16, Epoch=12, Total Steps=2144, Batch Size=4,

Optimizer = Prodigy.

Please support the original illustrator Xin&Obiwan:

X: https://twitter.com/moehime_union

Pixiv: https://www.pixiv.net/users/258003

Fanbox: https://moehime.fanbox.cc

BTW you can support the model author Jack Owlfelino as well :)

X: https://twitter.com/JackOwlfelino

Pixiv: https://www.pixiv.net/users/103422885

Works well on KohakuXL Delta & AnimagineXL V3.0. The compatibility on other checkpoints has not been verified yet.

KohakuXL Delta: https://civarchive.com/models/332076

AnimagineXL V3.0: https://civarchive.com/models/260267

Best weight range: 0.6~1.0

CFG scale: 5~7

Width×Height: 960x1440

Trigger prompt: obiwan,

再次强调画风模型只是对社区模型对某些特定画风的模拟效果不佳的弥补，这东西是给普通玩家带来快乐而不是拿来给画师跳脸添堵的。Xin姐和她的萌姬联合社团是我一直以来都很喜欢的画师，请尊重并支持人类画师的劳动成果，如当事方提出版权诉求将即刻删除模型，同时保留追究违规使用生成图像者责任的权利。