https://civitai.com/models/12427/old-school-shoujo
权重150 Weight 150
https://civitai.com/models/6304?modelVersionId=7387
权重650 Weight 650
原计划上传一个训练错误的LORA混合样本，不过这个组合也很不错。
Noob-leyon LORA训练时出现了错误，我会上传一个图片进行演示。这个现象其实很常见，大部分的LORA都不满足混合的最低标准，据我观察合格率大概有百分之一。
以下是混合中LORA的常见表现
LORA未完整学习，LORA学习深度不足，LORA错误学习。还有一部分LORA训练标注有问题，总之，早期LORA存在很多很多的不足，我希望现在的LORA已经有所改善。
但是在那之前我需要想办法解决提示词标注同类化造成的混合稀释，嗯看起来完全不可能做到。
I originally planned to upload a training error lora mixed sample, but this combination is also quite good
noob - an error occurred during the training of leyon lora, and I will upload an image to demonstrate. This phenomenon is actually quite common; most lora do not meet the minimum standards for mixing. From my observations, the pass rate is about 1 percent.
Here are the common manifestations of mixed Lora.
Lora has not fully learned, the depth of Lora's learning is insufficient, and there are errors in Lora's learning. Additionally, some of the training annotations for Lora have issues. In short, there were many shortcomings in the early stages of Lora, and I hope that the current version of Lora has improved.
But before that, I need to find a way to address the mixed dilution caused by the homogenization of prompt labeling, which seems completely impossible to achieve
