pretrained model on FFHQ for https://github.com/yoctta/sd_personalization_encoder.