Type | |
Stats | 330 107 |
Reviews | (26) |
Published | Mar 13, 2024 |
Base Model | |
Trigger Words | datou |
Hash | AutoV2 73559FDD5B |
【模型简介】
LoRA训练集由各种大头效果的漫画风格组成,算早期的真实人物夸张效果的漫画头像。结合人像类大模型可以稳定生成写实,手绘,厚涂或2.5D等风格的人物半身照。模型泛化测试全面,发色,服饰,配饰,瞳色等均可自由替换;
【模型出图参数】
大模型:AWpainting/麦橘写实/ChilloutMix等等(人像,绘画类的均可)
CLIP跳过层:1-2
VAE:按需选择
采样方法:Eulera
迭代步数:18-22步
推荐出图尺寸:806*1120(不同大模型下可能出现多肢体或多头,可调整分辨率大小解决)
提示词引导系数:7.0
【出图关键词】
触发词:datou
正向提示词(参考):
1boy,male focus,solo,necktie,formal,belt,shirt,blue necktie,smile,suit,realistic,jacket,pants,brown hair,looking at viewer,white shirt,vest,black jacket,simple background,hands in pockets,collared shirt,black pants,brown eyes,teeth,grin,short hair,black suit,long sleeves,grey background,facial hair,datou,(best quality:1.5), (masterpiece:1.5), (super Realistic:1.5),(highdetail:1.5)
负面提示词:
(worst quality, low quality:2),(normal quality:1.5),monochrome,((watermark:1.4)),artist name,character name,text,bad anatomy,sketch,duplicate,(ng_deepnegative_v1_75t),negative_hand-neg,bad_pictures,EasyNegative,