from diffsynth.pipelines.qwen_image_21 import QwenImage21Pipeline, ModelConfig, ControlNetInput from modelscope import dataset_snapshot_download from PIL import Image import torch from diffsynth.core.quant import QuantizeConfig vram_config = { "offload_dtype": torch.bfloat16, "offload_device": "cpu", "onload_dtype": torch.bfloat16, "onload_device": "cpu", "preparing_dtype": torch.bfloat16, "preparing_device": "cuda", "computation_dtype": torch.bfloat16, "computation_device": "cuda", "quantize": QuantizeConfig(method="entropack_lossy_quant_int8_raw", exclude_modules=["control_img_in"]), } pipe = QwenImage21Pipeline.from_pretrained( torch_dtype=torch.bfloat16, device="cuda", model_configs=[ ModelConfig(model_id="Qwen/Qwen-Image-2.1", origin_file_pattern="transformer/diffusion_pytorch_model*.safetensors", **vram_config), ModelConfig(model_id="Qwen/Qwen-Image-2.1", origin_file_pattern="text_encoder/model*.safetensors", **vram_config), ModelConfig(model_id="Qwen/Qwen-Image-2.1", origin_file_pattern="vae/diffusion_pytorch_model*.safetensors", **vram_config), ModelConfig(model_id="PAI/Qwen-Image-2.1-Fun-Controlnet-Union", origin_file_pattern="Qwen-Image-2.1-Fun-Controlnet-Union.safetensors", **vram_config), ], processor_config=ModelConfig(model_id="Qwen/Qwen-Image-2.1", origin_file_pattern="processor/"), vram_limit=torch.cuda.mem_get_info("cuda")[1] / (1024 ** 3) - 0.5, ) dataset_snapshot_download( dataset_id="DiffSynth-Studio/diffsynth_example_dataset", local_dir="./data/diffsynth_example_dataset", allow_file_pattern="qwen_image_21/Qwen-Image-2.1-Fun-Controlnet-Union/*", ) # Control (canny / depth / grayscale / hed / lineart / mlsd / pose / scribble) dataset_snapshot_download( dataset_id="DiffSynth-Studio/example_image_dataset", local_dir="./data/example_image_dataset", allow_file_pattern="depth/image_1.jpg" ) controlnet_image = Image.open("data/example_image_dataset/depth/image_1.jpg").resize((1024, 1024)) prompt = "精致肖像,水下少女,蓝裙飘逸,发丝轻扬,光影透澈,气泡环绕,面容恬静,细节精致,梦幻唯美。" image = pipe( prompt=prompt, seed=0, height=1024, width=1024, controlnet_inputs=[ControlNetInput(image=controlnet_image, scale=0.7)], num_inference_steps=40, ) image.save("image_control.png") # Inpaint dataset_snapshot_download( dataset_id="DiffSynth-Studio/example_image_dataset", local_dir="./data/example_image_dataset", allow_file_pattern="inpaint/*.jpg" ) inpaint_image = Image.open("./data/example_image_dataset/inpaint/image_1.jpg").convert("RGB").resize((1024, 1024)) inpaint_mask = Image.open("./data/example_image_dataset/inpaint/mask.jpg").convert("RGB").resize((1024, 1024)) prompt = "一只戴着墨镜的猫" image = pipe( prompt=prompt, seed=0, height=1024, width=1024, controlnet_inputs=[ControlNetInput(inpaint_image=inpaint_image, inpaint_mask=inpaint_mask, scale=0.7)], num_inference_steps=40, ) image.save("image_inpaint.png")