V2 Update

1.进行了数据集的精选！除去了画师的摸鱼图
2.从一个论文抄袭了一些未经证实的东西，理论上应该能提升构图水平和在低步数生成时的稳定性

1. Curated the dataset! Removed the artist's quick doodles.

2. Borrowed some unverified methods from a research paper; theoretically, they should improve composition and stability at low step.

这个版本选用了epoch16，但实际上有其他候选ep，最后选择16的原因是因为画的脸最好看 这里是其他ep的链接
This version uses Epoch 16. There were other candidate epochs, but I ended up choosing 16 because it generates the best-looking faces. Here are the links to the other epochs:

ep22:https://drive.google.com/file/d/1s9RX6BuVPQWRI0dHwCT92VwRAVVE1OHp/view?usp=sharing
ep20:https://drive.google.com/file/d/1-_0e3hJiGaPVp4Ge_IF6yJjd98pCe7PS/view?usp=sharing
ep18:https://drive.google.com/file/d/1JJvOnGH17Dn9cG7NpY89S6SpeNJDff1p/view?usp=sharing




Trained on Anima Base 1.0