I couldn't generate images properly after training the model with Dreambooth.What did I do wrong?

[huangxuzhong]

I first ran train_scripts/train_dreambooth.py, and the training process was completed normally. Then I used tools/convert_pixart_alpha_to_diffusers.py to transform the model, and this step was also successfully completed. However, during the prediction, no matter what prompt words were used, the following images were generated. What should I do?

[huangxuzhong]

I have solved this problem. The reason for this issue is that I mistakenly removed this line of code: load_from=""
But I have another question, how can I directly load and save diffusers version checkpoints instead of .pth(Do I have to use convert_pixart_alpha_to_diffusers.py convert format?)