ver.3.0 face more accurate, but i accidentally train with blurry photo, so the weight suggest in 0.7, if too high, the photo will getting blurry

and also you can generate without trigger word, just need to mention japanese, other wise the face doesn't look like asian.

3.0版本脸训练到最像，但是因为训练的时候不小心用到带有景深效果的照片，所以权重最好是在0.7以下，要是到了0.8就会出现模糊, 还有这个模型也可以不需要任何关键字，但是一些模型会出现美国人的脸，所以要在prompt注明japanese。