Use regularized training datasets. Bigger and richness in the background.










It's a retrained Locon model which should work better with clothes. It works better with the anime base model. This LoRA needs to be used in with negative embedding models such as EasyNegative, Deep Negative, etc.
Two separate clothing tags can be used:
skirt: do not use other skirt modifier tags (like red skirt, long skirt), or they may not be recognized.
dress, dress lift: other dress modifier tags may also reduce recognition.
Add tags like nude if you don't want clothes.
Suggestions for realistic base models:
Add stain ribbon tag to stabilize the ribbon material.
Added negative tag (rope:1.8) to remove rope texture.
Reduce the Hires. fix denoising strength to avoid distortion (around 0.52).
You can check out my example images to use. Unfortunately, I know that this LoRA tends to produce strange and distorted output on many base models.
Use regularized training datasets. Bigger and richness in the background.
1. 轉载模型僅供學習與交流分享,其版權及最終解释權归原作者。
2. 模型原作者如需認領模型,請通過官方渠道联系海藝AI工作人員進行認證。我們致力於保護每一位創作者的權益。 點擊去認領
