推荐使用NAI模型和easynegative
训练集使用了227张图，在 batch size = 4 下训练。unet表现尚可，text encoder表现的不是很理想，可能是学习率太低引起的欠拟合（所以调batch size稍微有点玄学）
希望喜欢
Recommend using the NAI model and easynegative embedding.
227 images were used in training set and trained under the parameter batch size=4. The unet performance is excellent, but text encoder is not very well. It may due to underfitting caused by low learning rate.
Hoping you enjoy it.