在krea2 raw (int 4 convrot)上进行训练,如果追求NSFW,建议使用pornmasterKrea2 v2 int4 convrot微调模型,不要nsfw就使用官方的krea2 turbo int4 convrot模型。
图片比例最好为竖图,横图尽量不要使用,要使用比列也不能大于3:2
v2.0:
训练数据集中添加大量由gpt image 2.5生成的各式各样的衣着打扮。该版本主要是学习婉儿的脸部特征,主要参考的脸是CG动画中最经典的一张截图。v2.0适合给婉儿换装,同时换装的婉儿身材会比原装更加丰满一些。
触发词:shangguanwaner, a woman, dark brown high ponytail, thick blunt bangs, narrow almond-shaped dark brown eyes, cool fair skin, oval face, golden twin-horn hairpiece with blue tassels.
如果要穿上原装衣服,就加入white-gold wide-sleeve form-fitting bodysuit, black thigh-high boots with gold and blue accents.
V1.0:
训练的图片取自TX的《王者荣耀剧场版》CG。
触发词:
shangguanwaner, a woman, long black hair in a high ponytail, narrow almond-shaped dark brown eyes, Chinese-style tight outfit, black and gold tall boots.
强度建议:0.8-1
Description
训练数据集中添加大量由gpt image 2.5生成的各式各样的衣着打扮。该版本主要是学习婉儿的脸部特征,主要参考的脸是CG动画中最经典的一张截图。v2.0适合给婉儿换装,同时换装的婉儿身材会比原装更加丰满一些。
触发词:shangguanwaner, a woman, dark brown high ponytail, thick blunt bangs, narrow almond-shaped dark brown eyes, cool fair skin, oval face, golden twin-horn hairpiece with blue tassels.
如果要加原装衣服,就加入white-gold wide-sleeve form-fitting bodysuit, black thigh-high boots with gold and blue accents.
强度:0.8-1
FAQ
Comments (1)
希望能保留动作提示词



















