Another style LoRA trained with only seven images, which means a certain degree of overfitting is unavoidable. Follow these usage tips:
1. Use DPM++ 2M sampler, Euler or Euler_a (these are tested) as I found that some samplers were not working well
2. Specify character appearance to avoid identical faces
3.Try a few more times, because hands and feet are not very stable.
Aside from that, this is an exceptionally good style LoRA (because the original illustrations are so good), definitely worth a try.
---
又是一个仅用七张图像训练的 LoRA（居然效果也是还行），不过过拟合还是不可避免。使用须知：
1. 使用 DPM++ 2M 采样器、euler 或 euler_a 采样器（已测试），因为我发现某些采样器效果不好
2. 为了避免总是生成相同的角色，一定需要指定角色外貌
3. 多试几次，因为手和脚的生成并不稳定
除此之外，这个 LoRA 效果挺好（因为原作画得太 TM 好了）。
（另外，我发现这种 LoRA 还有让任意角色身材都变得更好的附带功效）