V3更新：
基于anythingV5底模制作，例图使用mix-pro4直出，大幅度改善了过拟合、换装难和娘化的问题
使用方法：
因为魈的角色特征较为复杂，还原原版服装，推荐权重0.8~0.9，触发词为：1boy, male focus, yellow eyes, (official:1.2)
换装则推荐较低权重0.6~0.8，并拉高服装权重，触发词为：1boy, male focus, yellow eyes
translate by GPT：
V3 Update:
Based on the anythingV5 base model, the example images are output using mix-pro4, which greatly improves the problems of overfitting, difficulty in changing costumes, and feminization.
How to use:
Due to the complex character features of Xiao, it is recommended to use a weight of 0.8-0.9 to restore the original costume, with trigger words：1boy, male focus, yellow eyes, (official:1.2).
For changing costumes, it is recommended to use a lower weight of 0.6-0.8 and increase the weight of the clothing, with trigger words：1boy, male focus, yellow eyes
Description
增添了训练集和训练方法，将训练底模更换为AOM3
增加了触发词，角色特征更加准确
Added more training data and changed the training method, replacing the base model with AOM3.
Added trigger words to improve the accuracy of character features.
FAQ
Comments (4)
希望lora模型做香菱，神里绫华以及其他原神角色的models,（就是你做类似的lora）
请问lora的训练集你用的是什么样的图啊？我自己训练出来总是面部感觉有些崩，没那么清晰
15张官图，5张同人，10张模型截图来的，我的模型出全身也容易崩脸就是
天哪我搞出来一张超好看的！
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.