According to the Lora model created based on the characters in the Tokyo Afterschool Summoners game, it is recommended to keep the training stability between 0.6 and 0.7 for the first attempt.
Inputting clothing directly generates Christmas costumes.
It can be used to create both SFW and NSFW content.
Description
重新训练后的lora，存储占用更小更稳定，
基于Indigo Furry mix，写实、漫画、混合均可使用
第一次练的版本1问题很多，这次解决了脸崩和跑偏的问题，同时降低了对背景的污染，推荐0.65-0.8，更高的权重还没来得及试。
Based on the Indigo Furry mix, the first version for practicing allows the use of realistic, comic, and hybrid styles. It had many issues, such as face distortion and drifting. However, these problems have been solved in this iteration, and background contamination has been reduced. We recommend using a weight range of 0.65-0.8, as higher weights have not been tested yet.