import torch from diffsynth.pipelines.moss_image import MOSSImagePipeline, ModelConfig from diffsynth.core import load_state_dict pipe = MOSSImagePipeline.from_pretrained( torch_dtype=torch.bfloat16, device="cuda", model_configs=[ ModelConfig(model_id="openmoss/MOSS-Image-Base", origin_file_pattern="transformer/diffusion_pytorch_model*.safetensors"), ModelConfig(model_id="openmoss/MOSS-Image-Base", origin_file_pattern="text_encoder/model*.safetensors"), ModelConfig(model_id="openmoss/MOSS-Image-Base", origin_file_pattern="vae/diffusion_pytorch_model.safetensors", computation_dtype=torch.float32), ], tokenizer_config=ModelConfig(model_id="openmoss/MOSS-Image-Base", origin_file_pattern="tokenizer/"), ) state_dict = load_state_dict("./models/train/MOSS-Image-Base_full/epoch-1.safetensors") pipe.dit.load_state_dict(state_dict, strict=False) prompt = "a dog" image = pipe(prompt, height=1024, width=1024, seed=0, num_inference_steps=50, cfg_scale=3.0) image.save("image_full.jpg")