这是一个仅用ke-ta的一张魂魄妖梦的图片训练的lora，它被训练来画ke-ta风格的魂魄妖梦。它基本继承了ke-ta风格的魂魄妖梦，而且过拟合不那么严重。此外令我惊奇的是的它能把眼睛画的非常清澈透亮。它的问题是它会把画面画的亮度过高，有谁能解决亮度过高的问题请告诉我。推荐的prompt是konpaku youmu。
（translate by chatgpt3.5）This is a Lora model trained solely on a single image of ke-ta's Youmu Konpaku. Its purpose is to draw Youmu Konpaku in the distinctive style of ke-ta. It successfully captures the essence of ke-ta's interpretation of Youmu Konpaku and doesn't suffer from severe overfitting. One remarkable aspect is its ability to depict the eyes with exceptional clarity and radiance. However, it tends to produce images with excessive brightness. If anyone knows how to address this issue of excessive brightness, please share your solution. I recommend using "Konpaku Youmu" as the suggested prompt.
=======================================================
过亮解决方案：Miyuutsu提到了使用Brightness Tweaker LoRA可以解决亮度问题。
Solution for overly bright display: Miyuutsu suggested using Brightness Tweaker LoRA to solve the brightness issue.
Description
FAQ
Comments (3)
While it won't fix what appears to be some over-fitting, you can always try merging the Brightness Tweaker LoRA with a negative value or recommend people use it. Best recommendation for the future is to save multiple epochs since a few epochs less likely would have made the biggest difference.
可以通过图像编辑软件修正原版素材的亮度，再重新训练
训练的方法是特殊的，并不是直接把一张图片重复训练9600次，那样lora不会有现在展示的效果。
The training method is unique - it's not simply repeating the same image 9600 times. That's because such an approach wouldn't produce the impressive results we see with LoRA now.