import torch from PIL import Image from safetensors.torch import load_file from diffsynth.pipelines.sensenova_u1_image import SenseNovaU1ImagePipeline, ModelConfig pipe = SenseNovaU1ImagePipeline.from_pretrained( torch_dtype=torch.bfloat16, device="cuda", model_configs=[ ModelConfig(model_id="SenseNova/SenseNova-U1.5-8B-MoT-SFT", origin_file_pattern="model*.safetensors"), ], tokenizer_config=ModelConfig(model_id="SenseNova/SenseNova-U1.5-8B-MoT-SFT", origin_file_pattern="./"), ) state_dict = load_file("./models/train/SenseNova-U1.5-8B-MoT-SFT-Edit_full/epoch-1.safetensors", device="cuda") pipe.dit.load_state_dict(state_dict, strict=False) edit_image = Image.open("data/diffsynth_example_dataset/sensenova_u1/SenseNova-U1.5-8B-MoT-SFT-Edit/edit/image1.jpg").convert("RGB") image = pipe( prompt="将裙子改为粉色", edit_image=edit_image, seed=42, height=1024, width=1024, num_inference_steps=50, cfg_scale=4.0, shift=3.0, ) image.save("image.jpg")