This is my first LoRA.Designed to generate black legwears on the left and white legwear on the right for your character.Sometimes AI may not tell out left and right.

It includes two kinds:blackwhitethighhighs and blackwhitepantyhose.If you want to see zettai ryouiki(namely abosolute territory,the uncovered area between skirt and thighhighs),please use the former trigger word and add "zettai ryouiki" in prompts.

Trigger words:blackwhitethighhighs,blackwhitepantyhose (use one at a time is ok)

There's no precise recommandation on weights,but according to my test,it can work with weight as low as 0.5 on some models.If you find it hard to strip off all the clothes you can try low weights or change a sampler.

It's trained on NAI-anime,so I don't know how it works with realistic models.

Some training paramaters:dimension=22,max step=2134,epoch=20.

Recommended to use with EasyNegative.

Before knowing proper parameters in training I tried dimension=64 and 12000 steps on such simple a concept,and unsurprisingly get an overfitting model...

这是我的第一个LoRA。目的在于让你的角色左腿穿黑丝右腿穿白丝。有时AI可能分不清左右。

模型内包含两个款式：黑白过膝袜和黑白连裤袜。如果你想要看绝对领域（也就是过膝袜和裙子之间露出来的那一块）请使用前一个触发词并在提示词内加入"zettai ryouiki"。

触发词：blackwhitethighhighs,blackwhitepantyhose（一次用一个就行）

没有准确的权重推荐，不过据我的测试对于一些底模即使在0.5的权重下它也能生效。如果你很难脱光角色的所有衣服，可以试试较低的权重或者换个采样方法。

基于NAI二次元底模训练，没试过在真实系模型上表现怎么样。

训练的dim值为22，最大步数2134，20个epoch。

推荐搭配EasyNegative。

刚开始炼丹的时候对参数还没底，对这么简单的概念一开始用了64的dim训了12000步，不出意外地过拟合了...