First,You can still use the previous Lora,This lora is not perfect,Although I don't have a better idea at presentCompared with the previous Lora,this lora added more samples and replaced the model,now this lora has a higher upper limit ,But still unstable,Not recommended and others face lora（like Korea doll）,lora proportion can refer to my uploaded pictures,Please add it yourself tag ,so as to solve possible problems(like nasolabial folds、so big eyes) ,Finally,please roll more pictures,When luck is bad,Ten times can't produce good results,首先，你依旧可以使用上一个lora，因为这个2.0并不是完美无缺的，所以才叫beta,但目前我确实想不到别的办法来提升了。比起上一个random coser face，添加了更多的样本以及更换底模，从而进一步稀释了长相特征和优化模型效果，优点是得到了更高的上限，缺点是依旧不稳定，还是必须进行一定数量的roll图，权重比可以参考我的图片，具体可以根据个人需求进行调整，不推荐和其他长相的lora混用（比如Korea doll之类的），建议搭配chilloutmix一起使用,最后就是多多出图，倒霉的时候连续10张都有瑕疵也是很正常的。

Files with SHA256: bb7ffe0556246266c66c070be194d90c414ca35f47d5ec98f7131e95cb8d1c84