V2 update!

v2采用了更原始的图像，没有去除水印，所以没有去水印带来的训练集质量的下降，在此之前去除了部分训练集里的色差效果，并且精选了训练集，所以可以生成更高质量的图像！（大概）以及可以召唤画师标志性的大水印！
V2 uses more raw images without the watermarks, avoiding the quality degradation in the training set caused by watermark removal. Prior to this, chromatic aberration was from parts of the dataset, and the training set was carefully curated. Therefore, it can generate higher-quality images! (Probably) Plus, you can now summon the artist's signature giant watermark!



trained on base1.0