自己训练的一个人脸 Lora 模型，基于 SD1.5，算是失败品，因为生成的人脸并不像用来训练的人脸，但是生成的人脸看起来也不错，于是分享出来给大家试试效果。

与 majicMIX realistic 或 majicMIX sombre 搭配使用，推荐参数：

模型权重: 0.5~0.9

提示词引导系数 CFG Scale: 5~7

采样步数 Sampling steps: 15~25

可以通过分层控制减少污染，例如：1,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0，分层控制可以尝试更高的模型权重。

The Lora model of a person's face that I trained myself, based on SD1.5, is considered a failure because the generated face does not look like the one used for training, but the generated face also looks good. So I shared it with everyone to try the effect.

Used in conjunction with majicMIX realistic or majicMIX symbol.

Pollution can be reduced through hierarchical control, such as 1,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0, Hierarchical control can attempt higher model weights.