Because of the training process for models with diffusion model (details about diffusion model is in the paper) is not stable, we update the training process for a more stable process. Raw training is ...