训练集来源于网络爬取图片,使用时请参考我的提示词。
The training set is from the network crawl picture, please refer to my prompt words when using.
底模chilloutmix_Ni
The base-model is chilloutmix_Ni.
qincai,1girl即为触发词。
qincai and 1girl were trigger words.
本lora的特点是头发好看,光影感真实,输出的时候脸部会占比较大,想出全身图可以把full body权重调高。
The features of this LoRa are beautiful hair and real sense of light and shadow. When output, the face will take up a large proportion, and the weight of the full body can be adjusted higher by coming up with the whole body picture.
提示词中的“ulzzang-6500”或者“pureerosface”为文本反转模型,请自行下载。
"ulzzang-6500" or "pureerosface" in the prompt word is a text reversal model. Please download it yourself.
全身图容易崩脸,使用SwinIR_4x高清修复即可,如果你没有这个选项,可以在WEBUI-选项-Stable Diffusion里面勾选上“Upcast cross attention layer to float32”(在页面最底端)
Full body pictures are easy to crack,Use SwinIR_4x HD to fix it. If you don't have this option, check "Upcast cross attention layer to float32" in webu-Options-stable Diffusion.
这是本人的第二个三次元真人lora,如果觉得效果可以,请给我一个关注和赞。
This is my second three-dimensional real person lora. If you think the effect is OK, please give me your subscribe and praise.
需要我帮忙练三次元lora也可以联系我。
Please contact me if you need my help to train a 3D real person LoRa