Update:

v2.0 is trained using generated images from v1.0 and it has better understanding of tags like [pantyhose] or even [nude]. Also, compared with v1.0, v2.0 is not easy to make images distorted.

This LoRA is able to generate the upskirt view of a girl.

The recommended weight is 0.5~1.0.

It dose not well in generating high resolution images, but you can use Hires. fix.

You can also generate the front view by yourself and make these 2 images together (Sample images with 2 views are all made in this way instead of generating directly).

v2.0使用v1.0生成图二次训练，泛化性有很大增强，对[pantyhose],[nude]等tag的理解加深，并且高权重下图片损坏的概率降低很多。

这个LoRA可以用来生成裙底图片。

推荐权重：0.5~1.0。

它不太擅长出大图，如果需要高清图片，推荐用高分辨率修复

你也可以自己另外出一张正面照，然后把它和裙底照拼成一张图（展示的双视角图片都是这样弄的，不是直接生成）