Trigger words/关键词：anranWeight/权重：0.7-0.9Resolution recommendation/分辨率推荐：512×768The first time I tried to use regularization for real lora training, the generalization was significantly improved compared with before, and the training efficiency was also improved a lot. Suggestions and comments are welcome第一次尝试启用正则化进行真人lora训练，泛化性对比之前有明显提升，训练的效率也提升了许多。欢迎提供建议与意见进行学习交流。Suggestion：using the ADetailer！建议：使用ADetailerAttention： this lora just for learning. Please respect public figures. Don’t share anything about NSFW.注意：这个lora只是为了学习，请尊重公众人物。请勿分享任何关于NSFW的内容。

安然 Anran

一个简单的模型a simple model

Files with SHA256: 980c7b9025d48d29f98744847820e34172729bb628969d3a663985b2758afcc4