new app.py

app.py

@@ -4,337 +4,614 @@ import sys
 from typing import Sequence, Mapping, Any, Union
 import torch
 import gradio as gr
-from PIL import Image
-from huggingface_hub import hf_hub_download
-import spaces
-from comfy import model_management
-
-hf_hub_download(repo_id="black-forest-labs/FLUX.1-Redux-dev", filename="flux1-redux-dev.safetensors", local_dir="models/style_models")
-hf_hub_download(repo_id="black-forest-labs/FLUX.1-Depth-dev", filename="flux1-depth-dev.safetensors", local_dir="models/diffusion_models")
-hf_hub_download(repo_id="Comfy-Org/sigclip_vision_384", filename="sigclip_vision_patch14_384.safetensors", local_dir="models/clip_vision")
-hf_hub_download(repo_id="Kijai/DepthAnythingV2-safetensors", filename="depth_anything_v2_vitl_fp32.safetensors", local_dir="models/depthanything")
-hf_hub_download(repo_id="black-forest-labs/FLUX.1-dev", filename="ae.safetensors", local_dir="models/vae/FLUX1")
-hf_hub_download(repo_id="comfyanonymous/flux_text_encoders", filename="clip_l.safetensors", local_dir="models/text_encoders")
-t5_path = hf_hub_download(repo_id="comfyanonymous/flux_text_encoders", filename="t5xxl_fp16.safetensors", local_dir="models/text_encoders/t5")
-
-# Import all the necessary functions from the original script
+
+
 def get_value_at_index(obj: Union[Sequence, Mapping], index: int) -> Any:
+    """Returns the value at the given index of a sequence or mapping.
+
+    If the object is a sequence (like list or string), returns the value at the given index.
+    If the object is a mapping (like a dictionary), returns the value at the index-th key.
+
+    Some return a dictionary, in these cases, we look for the "results" key
+
+    Args:
+        obj (Union[Sequence, Mapping]): The object to retrieve the value from.
+        index (int): The index of the value to retrieve.
+
+    Returns:
+        Any: The value at the given index.
+
+    Raises:
+        IndexError: If the index is out of bounds for the object and the object is not a mapping.
+    """
     try:
         return obj[index]
     except KeyError:
         return obj["result"][index]
 
-# Add all the necessary setup functions from the original script
+
 def find_path(name: str, path: str = None) -> str:
+    """
+    Recursively looks at parent folders starting from the given path until it finds the given name.
+    Returns the path as a Path object if found, or None otherwise.
+    """
+    # If no path is given, use the current working directory
     if path is None:
         path = os.getcwd()
+
+    # Check if the current directory contains the name
     if name in os.listdir(path):
         path_name = os.path.join(path, name)
         print(f"{name} found: {path_name}")
         return path_name
+
+    # Get the parent directory
     parent_directory = os.path.dirname(path)
+
+    # If the parent directory is the same as the current directory, we've reached the root and stop the search
     if parent_directory == path:
         return None
+
+    # Recursively call the function with the parent directory
     return find_path(name, parent_directory)
 
+
 def add_comfyui_directory_to_sys_path() -> None:
+    """
+    Add 'ComfyUI' to the sys.path
+    """
     comfyui_path = find_path("ComfyUI")
     if comfyui_path is not None and os.path.isdir(comfyui_path):
         sys.path.append(comfyui_path)
         print(f"'{comfyui_path}' added to sys.path")
 
+
 def add_extra_model_paths() -> None:
+    """
+    Parse the optional extra_model_paths.yaml file and add the parsed paths to the sys.path.
+    """
     try:
         from main import load_extra_path_config
     except ImportError:
+        print(
+            "Could not import load_extra_path_config from main.py. Looking in utils.extra_config instead."
+        )
         from utils.extra_config import load_extra_path_config
+
     extra_model_paths = find_path("extra_model_paths.yaml")
+
     if extra_model_paths is not None:
         load_extra_path_config(extra_model_paths)
     else:
         print("Could not find the extra_model_paths config file.")
 
-# Initialize paths
+
 add_comfyui_directory_to_sys_path()
 add_extra_model_paths()
 
+
+# MODIFIED FUNCTION - THE CORE FIX IS HERE
 def import_custom_nodes() -> None:
+    """
+    This function now correctly mimics the necessary parts of ComfyUI's startup
+    to ensure all paths and nodes are initialized.
+    """
     import asyncio
     import execution
+
+    
+    # Crucially, import the main module to access its functions
+    import main as comfyui_main 
     from nodes import init_extra_nodes
     import server
+    # 1. Apply paths from extra_model_paths.yaml and command-line args (if any)
+    # This is the step that was missing and caused the 'model_paths' error.
+    comfyui_main.apply_custom_paths()
+    print("Applied custom paths from extra_model_paths.yaml")
+
+    # 2. Initialize the server and queue (needed as a dependency for some nodes)
+    # We create a new loop each time, as per the original request to keep logic inside the function.
     loop = asyncio.new_event_loop()
     asyncio.set_event_loop(loop)
     server_instance = server.PromptServer(loop)
     execution.PromptQueue(server_instance)
+
+    # 3. Initialize the custom nodes. This will now work because paths are set.
     init_extra_nodes()
+    print("Custom nodes initialized.")
 
-# Import all necessary nodes
-from nodes import (
-    StyleModelLoader,
-    VAEEncode,
-    NODE_CLASS_MAPPINGS,
-    LoadImage,
-    CLIPVisionLoader,
-    SaveImage,
-    VAELoader,
-    CLIPVisionEncode,
-    DualCLIPLoader,
-    EmptyLatentImage,
-    VAEDecode,
-    UNETLoader,
-    CLIPTextEncode,
-)
 
-# Initialize all constant nodes and models in global context
+from nodes import NODE_CLASS_MAPPINGS
+
 import_custom_nodes()
 
-# Global variables for preloaded models and constants
-#with torch.inference_mode():
-    # Initialize constants
-intconstant = NODE_CLASS_MAPPINGS["INTConstant"]()
-CONST_1024 = intconstant.get_value(value=1024)
-
-# Load CLIP
-dualcliploader = DualCLIPLoader()
-CLIP_MODEL = dualcliploader.load_clip(
-    clip_name1="t5/t5xxl_fp16.safetensors",
-    clip_name2="clip_l.safetensors",
-    type="flux",
+
+if "Florence2ModelLoader" in NODE_CLASS_MAPPINGS:
+    print("Manually initializing Florence2ModelLoader.INPUT_TYPES() to populate model paths.")
+    florence_class = NODE_CLASS_MAPPINGS["Florence2ModelLoader"]
+    florence_class.INPUT_TYPES()
+# =========================================================================
+
+checkpointloadersimple = NODE_CLASS_MAPPINGS["CheckpointLoaderSimple"]()
+checkpointloadersimple_50 = checkpointloadersimple.load_checkpoint(
+    ckpt_name="SD1.5/dreamshaper_8.safetensors"
+)
+
+cliptextencode = NODE_CLASS_MAPPINGS["CLIPTextEncode"]()
+
+controlnetloader = NODE_CLASS_MAPPINGS["ControlNetLoader"]()
+controlnetloader_73 = controlnetloader.load_controlnet(
+    control_net_name="SD1.5/control_v11p_sd15_openpose.pth"
+)
+
+loadimage = NODE_CLASS_MAPPINGS["LoadImage"]()
+
+
+florence2modelloader = NODE_CLASS_MAPPINGS["Florence2ModelLoader"]()
+florence2modelloader_204 = florence2modelloader.loadmodel(
+    model="Florence-2-base",
+    precision="fp16",
+    attention="sdpa",
+    convert_to_safetensors=False,
 )
+florence2run = NODE_CLASS_MAPPINGS["Florence2Run"]()
 
-# Load VAE
-vaeloader = VAELoader()
-VAE_MODEL = vaeloader.load_vae(vae_name="FLUX1/ae.safetensors")
 
-# Load UNET
-unetloader = UNETLoader()
-UNET_MODEL = unetloader.load_unet(
-    unet_name="flux1-depth-dev.safetensors", weight_dtype="default"
+checkpointloadersimple_319 = checkpointloadersimple.load_checkpoint(
+    ckpt_name="SD1.5/dreamshaper_8Inpainting.safetensors"
 )
 
-# Load CLIP Vision
-clipvisionloader = CLIPVisionLoader()
-CLIP_VISION_MODEL = clipvisionloader.load_clip(
-    clip_name="sigclip_vision_patch14_384.safetensors"
+loraloader = NODE_CLASS_MAPPINGS["LoraLoader"]()
+loraloader_338 = loraloader.load_lora(
+    lora_name="add_detail.safetensors",
+    strength_model=1,
+    strength_clip=1,
+    model=get_value_at_index(checkpointloadersimple_319, 0),
+    clip=get_value_at_index(checkpointloadersimple_319, 1),
+)
+
+
+
+loraloader_353 = loraloader.load_lora(
+    lora_name="BaldifierW2.safetensors",
+    strength_model=2,
+    strength_clip=1,
+    model=get_value_at_index(loraloader_338, 0),
+    clip=get_value_at_index(loraloader_338, 1),
 )
 
-# Load Style Model
-stylemodelloader = StyleModelLoader()
-STYLE_MODEL = stylemodelloader.load_style_model(
-    style_model_name="flux1-redux-dev.safetensors"
+controlnetloader_389 = controlnetloader.load_controlnet(
+    control_net_name="SD1.5/control_v11p_sd15_openpose.pth"
 )
 
-# Initialize samplers
-ksamplerselect = NODE_CLASS_MAPPINGS["KSamplerSelect"]()
-SAMPLER = ksamplerselect.get_sampler(sampler_name="euler")
+dwpreprocessor = NODE_CLASS_MAPPINGS["DWPreprocessor"]()
+controlnetapplyadvanced = NODE_CLASS_MAPPINGS["ControlNetApplyAdvanced"]()
+
+layerutility_imagescalebyaspectratio_v2 = NODE_CLASS_MAPPINGS[
+    "LayerUtility: ImageScaleByAspectRatio V2"
+]()
+
+
+layermask_personmaskultra_v2 = NODE_CLASS_MAPPINGS[
+    "LayerMask: PersonMaskUltra V2"
+]()
+
+growmask = NODE_CLASS_MAPPINGS["GrowMask"]()
 
-# Initialize depth model
-cr_clip_input_switch = NODE_CLASS_MAPPINGS["CR Clip Input Switch"]()
-downloadandloaddepthanythingv2model = NODE_CLASS_MAPPINGS["DownloadAndLoadDepthAnythingV2Model"]()
-DEPTH_MODEL = downloadandloaddepthanythingv2model.loadmodel(
-    model="depth_anything_v2_vitl_fp32.safetensors"
+inpaintmodelconditioning = NODE_CLASS_MAPPINGS["InpaintModelConditioning"]()
+ksampler = NODE_CLASS_MAPPINGS["KSampler"]()
+
+
+vaedecode = NODE_CLASS_MAPPINGS["VAEDecode"]()
+vaeencode = NODE_CLASS_MAPPINGS["VAEEncode"]()
+faceanalysismodels = NODE_CLASS_MAPPINGS["FaceAnalysisModels"]()
+faceanalysismodels_506 = faceanalysismodels.load_models(
+    library="insightface", provider="CPU"
+)
+upscalemodelloader = NODE_CLASS_MAPPINGS["UpscaleModelLoader"]()
+upscalemodelloader_835 = upscalemodelloader.load_model(
+    model_name="4x-UltraSharp.pth"
 )
+ipadapterunifiedloader = NODE_CLASS_MAPPINGS["IPAdapterUnifiedLoader"]()
+ipadapteradvanced = NODE_CLASS_MAPPINGS["IPAdapterAdvanced"]()
+facesegmentation = NODE_CLASS_MAPPINGS["FaceSegmentation"]()
+layerutility_imageblend_v2 = NODE_CLASS_MAPPINGS[
+    "LayerUtility: ImageBlend V2"
+]()
+image_comparer_rgthree = NODE_CLASS_MAPPINGS["Image Comparer (rgthree)"]()
+saveimage = NODE_CLASS_MAPPINGS["SaveImage"]()
+imageupscalewithmodel = NODE_CLASS_MAPPINGS["ImageUpscaleWithModel"]()
+
+
+# def main():
+def generate_image(model_image, hairstyle_template_image):
+
+
+
 
-cliptextencode = CLIPTextEncode()
-loadimage = LoadImage()
-vaeencode = VAEEncode()
-fluxguidance = NODE_CLASS_MAPPINGS["FluxGuidance"]()
-instructpixtopixconditioning = NODE_CLASS_MAPPINGS["InstructPixToPixConditioning"]()
-clipvisionencode = CLIPVisionEncode()
-stylemodelapplyadvanced = NODE_CLASS_MAPPINGS["StyleModelApplyAdvanced"]()
-emptylatentimage = EmptyLatentImage()
-basicguider = NODE_CLASS_MAPPINGS["BasicGuider"]()
-basicscheduler = NODE_CLASS_MAPPINGS["BasicScheduler"]()        
-randomnoise = NODE_CLASS_MAPPINGS["RandomNoise"]()
-samplercustomadvanced = NODE_CLASS_MAPPINGS["SamplerCustomAdvanced"]()
-vaedecode = VAEDecode()
-cr_text = NODE_CLASS_MAPPINGS["CR Text"]()
-saveimage = SaveImage()
-getimagesizeandcount = NODE_CLASS_MAPPINGS["GetImageSizeAndCount"]()
-depthanything_v2 = NODE_CLASS_MAPPINGS["DepthAnything_V2"]()
-imageresize = NODE_CLASS_MAPPINGS["ImageResize+"]()
-
-model_loaders = [CLIP_MODEL, VAE_MODEL, UNET_MODEL, CLIP_VISION_MODEL]
-
-model_management.load_models_gpu([
-    loader[0].patcher if hasattr(loader[0], 'patcher') else loader[0] for loader in model_loaders
-])
-
-@spaces.GPU
-def generate_image(prompt, structure_image, style_image, depth_strength=15, style_strength=0.5) -> str:
-    """Main generation function that processes inputs and returns the path to the generated image."""
     with torch.inference_mode():
-        # Set up CLIP
-        clip_switch = cr_clip_input_switch.switch(
-            Input=1,
-            clip1=get_value_at_index(CLIP_MODEL, 0),
-            clip2=get_value_at_index(CLIP_MODEL, 0),
+        cliptextencode_52 = cliptextencode.encode(
+            text="multiple_hands, multiple_legs, multiple_girls\nlow quality, blurry, out of focus, distorted, unrealistic, extra limbs, missing limbs, deformed hands, deformed fingers, extra fingers, long neck, unnatural face, bad anatomy, bad proportions, poorly drawn face, poorly drawn eyes, asymmetrical eyes, extra eyes, extra head, floating objects, watermark, text, logo, jpeg artifacts, overexposed, underexposed, harsh lighting, bad posture, strange angles, unnatural expressions, oversaturated colors, messy hair, unrealistic skin texture, wrinkles inappropriately placed, incorrect shading, pixelation, complex background, busy background, detailed background, crowded scene, clutter, messy elements, unnecessary objects, overlapping objects, intricate patterns, vibrant colors, high contrast, graffiti, shadows, reflections, multiple layers, unrealistic lighting, overexposed areas.",
+            clip=get_value_at_index(checkpointloadersimple_50, 1),
         )
-        
-        # Encode text
-        text_encoded = cliptextencode.encode(
-            text=prompt,
-            clip=get_value_at_index(clip_switch, 0),
+
+
+
+
+        loadimage_144 = loadimage.load_image(image=hairstyle_template_image)
+
+
+
+        florence2run_203 = florence2run.encode(
+            text_input="",
+            task="more_detailed_caption",
+            fill_mask=True,
+            keep_model_loaded=False,
+            max_new_tokens=1024,
+            num_beams=3,
+            do_sample=True,
+            output_mask_select="",
+            seed=random.randint(1, 2**64),
+            image=get_value_at_index(loadimage_144, 0),
+            florence2_model=get_value_at_index(florence2modelloader_204, 0),
         )
-        empty_text = cliptextencode.encode(
-            text="",
-            clip=get_value_at_index(clip_switch, 0),
+
+        cliptextencode_188 = cliptextencode.encode(
+            text=get_value_at_index(florence2run_203, 2),
+            clip=get_value_at_index(checkpointloadersimple_50, 1),
         )
-        
-        # Process structure image
-        structure_img = loadimage.load_image(image=structure_image)
-        
-        # Resize image
-        resized_img = imageresize.execute(
-            width=get_value_at_index(CONST_1024, 0),
-            height=get_value_at_index(CONST_1024, 0),
-            interpolation="bicubic",
-            method="keep proportion",
-            condition="always",
-            multiple_of=16,
-            image=get_value_at_index(structure_img, 0),
+
+
+
+
+
+
+
+        cliptextencode_836 = cliptextencode.encode(
+            text=" Bald, no hair, small head, small head, nothing around, no light, no highlights, no sunlight，Smooth forehead，No wrinkles",
+            clip=get_value_at_index(loraloader_353, 1),
         )
-        
-        # Get image size
-        size_info = getimagesizeandcount.getsize(
-            image=get_value_at_index(resized_img, 0)
+
+        cliptextencode_321 = cliptextencode.encode(
+            text="wrinkles,Big forehead, big head, big back of the head，multiple_hands, multiple_legs, multiple_girls\nlow quality, blurry, out of focus, distorted, unrealistic, extra limbs, missing limbs, deformed hands, deformed fingers, extra fingers, long neck, unnatural face, bad anatomy, bad proportions, poorly drawn face, poorly drawn eyes, asymmetrical eyes, extra eyes, extra head, floating objects, watermark, text, logo, jpeg artifacts, overexposed, underexposed, harsh lighting, bad posture, strange angles, unnatural expressions, oversaturated colors, messy hair, unrealistic skin texture, wrinkles inappropriately placed, incorrect shading, pixelation, complex background, busy background, detailed background, crowded scene, clutter, messy elements, unnecessary objects, overlapping objects, intricate patterns, vibrant colors, high contrast, graffiti, shadows, reflections, multiple layers, unrealistic lighting, overexposed areas.",
+            clip=get_value_at_index(loraloader_353, 1),
         )
-        
-        # Encode VAE
-        vae_encoded = vaeencode.encode(
-            pixels=get_value_at_index(size_info, 0),
-            vae=get_value_at_index(VAE_MODEL, 0),
+
+
+        loadimage_317 = loadimage.load_image(image=model_image)
+
+
+        dwpreprocessor_390 = dwpreprocessor.estimate_pose(
+            detect_hand="enable",
+            detect_body="enable",
+            detect_face="enable",
+            resolution=768,
+            bbox_detector="yolox_l.onnx",
+            pose_estimator="dw-ll_ucoco_384_bs5.torchscript.pt",
+            scale_stick_for_xinsr_cn="disable",
+            image=get_value_at_index(loadimage_317, 0),
         )
-        
-        # Process depth
-        depth_processed = depthanything_v2.process(
-            da_model=get_value_at_index(DEPTH_MODEL, 0),
-            images=get_value_at_index(size_info, 0),
+
+
+        controlnetapplyadvanced_388 = controlnetapplyadvanced.apply_controlnet(
+            strength=1,
+            start_percent=0,
+            end_percent=1,
+            positive=get_value_at_index(cliptextencode_836, 0),
+            negative=get_value_at_index(cliptextencode_321, 0),
+            control_net=get_value_at_index(controlnetloader_389, 0),
+            image=get_value_at_index(dwpreprocessor_390, 0),
+            vae=get_value_at_index(checkpointloadersimple_319, 2),
         )
-        
-        # Apply Flux guidance
-        flux_guided = fluxguidance.append(
-            guidance=depth_strength,
-            conditioning=get_value_at_index(text_encoded, 0),
+
+
+        layerutility_imagescalebyaspectratio_v2_331 = (
+            layerutility_imagescalebyaspectratio_v2.image_scale_by_aspect_ratio(
+                aspect_ratio="original",
+                proportional_width=1,
+                proportional_height=1,
+                fit="letterbox",
+                method="lanczos",
+                round_to_multiple="8",
+                scale_to_side="longest",
+                scale_to_length=768,
+                background_color="#000000",
+                image=get_value_at_index(loadimage_317, 0),
+                mask=get_value_at_index(loadimage_317, 1),
+            )
         )
-        
-        # Process style image
-        style_img = loadimage.load_image(image=style_image)
-        
-        # Encode style with CLIP Vision
-        style_encoded = clipvisionencode.encode(
-            crop="center",
-            clip_vision=get_value_at_index(CLIP_VISION_MODEL, 0),
-            image=get_value_at_index(style_img, 0),
+
+
+        layermask_personmaskultra_v2_327 = (
+            layermask_personmaskultra_v2.person_mask_ultra_v2(
+                face=False,
+                hair=True,
+                body=False,
+                clothes=False,
+                accessories=False,
+                background=False,
+                confidence=0.4,
+                detail_method="VITMatte",
+                detail_erode=6,
+                detail_dilate=6,
+                black_point=0.01,
+                white_point=0.99,
+                process_detail=True,
+                device="cuda",
+                max_megapixels=2,
+                images=get_value_at_index(
+                    layerutility_imagescalebyaspectratio_v2_331, 0
+                ),
+            )
         )
-        
-        # Set up conditioning
-        conditioning = instructpixtopixconditioning.encode(
-            positive=get_value_at_index(flux_guided, 0),
-            negative=get_value_at_index(empty_text, 0),
-            vae=get_value_at_index(VAE_MODEL, 0),
-            pixels=get_value_at_index(depth_processed, 0),
+
+
+        growmask_502 = growmask.expand_mask(
+            expand=20,
+            tapered_corners=True,
+            mask=get_value_at_index(layermask_personmaskultra_v2_327, 1),
         )
-        
-        # Apply style
-        style_applied = stylemodelapplyadvanced.apply_stylemodel(
-            strength=style_strength,
-            conditioning=get_value_at_index(conditioning, 0),
-            style_model=get_value_at_index(STYLE_MODEL, 0),
-            clip_vision_output=get_value_at_index(style_encoded, 0),
+
+
+        inpaintmodelconditioning_330 = inpaintmodelconditioning.encode(
+            noise_mask=True,
+            positive=get_value_at_index(controlnetapplyadvanced_388, 0),
+            negative=get_value_at_index(controlnetapplyadvanced_388, 1),
+            vae=get_value_at_index(checkpointloadersimple_319, 2),
+            pixels=get_value_at_index(layerutility_imagescalebyaspectratio_v2_331, 0),
+            mask=get_value_at_index(growmask_502, 0),
         )
-        
-        # Set up empty latent
-        empty_latent = emptylatentimage.generate(
-            width=get_value_at_index(resized_img, 1),
-            height=get_value_at_index(resized_img, 2),
-            batch_size=1,
+
+
+        ksampler_318 = ksampler.sample(
+            seed=random.randint(1, 2**64),
+            steps=10,
+            cfg=2.5,
+            sampler_name="euler_ancestral",
+            scheduler="normal",
+            denoise=1,
+            model=get_value_at_index(loraloader_353, 0),
+            positive=get_value_at_index(inpaintmodelconditioning_330, 0),
+            negative=get_value_at_index(inpaintmodelconditioning_330, 1),
+            latent_image=get_value_at_index(inpaintmodelconditioning_330, 2),
         )
-        
-        # Set up guidance
-        guided = basicguider.get_guider(
-            model=get_value_at_index(UNET_MODEL, 0),
-            conditioning=get_value_at_index(style_applied, 0),
+
+
+        vaedecode_322 = vaedecode.decode(
+            samples=get_value_at_index(ksampler_318, 0),
+            vae=get_value_at_index(checkpointloadersimple_319, 2),
         )
-        
-        # Set up scheduler
-        schedule = basicscheduler.get_sigmas(
-            scheduler="simple",
-            steps=28,
+
+
+        vaeencode_191 = vaeencode.encode(
+            pixels=get_value_at_index(vaedecode_322, 0),
+            vae=get_value_at_index(checkpointloadersimple_50, 2),
+        )
+
+
+
+
+        faceanalysismodels_840 = faceanalysismodels.load_models(
+            library="insightface", provider="CUDA"
+        )
+
+
+
+        # for q in range(1):
+        ipadapterunifiedloader_90 = ipadapterunifiedloader.load_models(
+            preset="PLUS (high strength)",
+            model=get_value_at_index(checkpointloadersimple_50, 0),
+        )
+
+        layerutility_imagescalebyaspectratio_v2_187 = (
+            layerutility_imagescalebyaspectratio_v2.image_scale_by_aspect_ratio(
+                aspect_ratio="original",
+                proportional_width=132,
+                proportional_height=1,
+                fit="letterbox",
+                method="lanczos",
+                round_to_multiple="8",
+                scale_to_side="longest",
+                scale_to_length=768,
+                background_color="#000000",
+                image=get_value_at_index(loadimage_144, 0),
+            )
+        )
+
+        ipadapteradvanced_85 = ipadapteradvanced.apply_ipadapter(
+            weight=1,
+            weight_type="strong style transfer",
+            combine_embeds="concat",
+            start_at=0,
+            end_at=1,
+            embeds_scaling="V only",
+            model=get_value_at_index(ipadapterunifiedloader_90, 0),
+            ipadapter=get_value_at_index(ipadapterunifiedloader_90, 1),
+            image=get_value_at_index(
+                layerutility_imagescalebyaspectratio_v2_187, 0
+            ),
+        )
+
+        dwpreprocessor_72 = dwpreprocessor.estimate_pose(
+            detect_hand="enable",
+            detect_body="enable",
+            detect_face="enable",
+            resolution=1024,
+            bbox_detector="yolox_l.onnx",
+            pose_estimator="dw-ll_ucoco_384_bs5.torchscript.pt",
+            scale_stick_for_xinsr_cn="disable",
+            image=get_value_at_index(vaedecode_322, 0),
+        )
+
+        controlnetapplyadvanced_189 = controlnetapplyadvanced.apply_controlnet(
+            strength=1,
+            start_percent=0,
+            end_percent=1,
+            positive=get_value_at_index(cliptextencode_188, 0),
+            negative=get_value_at_index(cliptextencode_52, 0),
+            control_net=get_value_at_index(controlnetloader_73, 0),
+            image=get_value_at_index(dwpreprocessor_72, 0),
+            vae=get_value_at_index(checkpointloadersimple_50, 2),
+        )
+
+        ksampler_45 = ksampler.sample(
+            seed=random.randint(1, 2**64),
+            steps=15,
+            cfg=1,
+            sampler_name="dpmpp_2m",
+            scheduler="karras",
             denoise=1,
-            model=get_value_at_index(UNET_MODEL, 0),
+            model=get_value_at_index(ipadapteradvanced_85, 0),
+            positive=get_value_at_index(controlnetapplyadvanced_189, 0),
+            negative=get_value_at_index(controlnetapplyadvanced_189, 1),
+            latent_image=get_value_at_index(vaeencode_191, 0),
         )
-        
-        # Generate random noise
-        noise = randomnoise.get_noise(noise_seed=random.randint(1, 2**64))
-        
-        # Sample
-        sampled = samplercustomadvanced.sample(
-            noise=get_value_at_index(noise, 0),
-            guider=get_value_at_index(guided, 0),
-            sampler=get_value_at_index(SAMPLER, 0),
-            sigmas=get_value_at_index(schedule, 0),
-            latent_image=get_value_at_index(empty_latent, 0),
+
+        vaedecode_67 = vaedecode.decode(
+            samples=get_value_at_index(ksampler_45, 0),
+            vae=get_value_at_index(checkpointloadersimple_50, 2),
         )
-        
-        # Decode VAE
-        decoded = vaedecode.decode(
-            samples=get_value_at_index(sampled, 0),
-            vae=get_value_at_index(VAE_MODEL, 0),
+
+        layermask_personmaskultra_v2_192 = (
+            layermask_personmaskultra_v2.person_mask_ultra_v2(
+                face=False,
+                hair=True,
+                body=False,
+                clothes=False,
+                accessories=False,
+                background=False,
+                confidence=0.4,
+                detail_method="VITMatte",
+                detail_erode=6,
+                detail_dilate=6,
+                black_point=0.01,
+                white_point=0.99,
+                process_detail=True,
+                device="cuda",
+                max_megapixels=2,
+                images=get_value_at_index(vaedecode_67, 0),
+            )
         )
-        
-        # Save image
-        prefix = cr_text.text_multiline(text="Flux_BFL_Depth_Redux")
-        
-        saved = saveimage.save_images(
-            filename_prefix=get_value_at_index(prefix, 0),
-            images=get_value_at_index(decoded, 0),
+
+        facesegmentation_505 = facesegmentation.segment(
+            area="face+forehead (if available)",
+            grow=-5,
+            grow_tapered=False,
+            blur=41,
+            analysis_models=get_value_at_index(faceanalysismodels_506, 0),
+            image=get_value_at_index(
+                layerutility_imagescalebyaspectratio_v2_331, 0
+            ),
+        )
+
+        growmask_396 = growmask.expand_mask(
+            expand=0,
+            tapered_corners=True,
+            mask=get_value_at_index(facesegmentation_505, 0),
+        )
+
+        layerutility_imageblend_v2_399 = layerutility_imageblend_v2.image_blend_v2(
+            invert_mask=True,
+            blend_mode="normal",
+            opacity=100,
+            background_image=get_value_at_index(
+                layerutility_imagescalebyaspectratio_v2_331, 0
+            ),
+            layer_image=get_value_at_index(vaedecode_322, 0),
+            layer_mask=get_value_at_index(growmask_396, 0),
+        )
+
+        layerutility_imageblend_v2_314 = layerutility_imageblend_v2.image_blend_v2(
+            invert_mask=True,
+            blend_mode="normal",
+            opacity=100,
+            background_image=get_value_at_index(layerutility_imageblend_v2_399, 0),
+            layer_image=get_value_at_index(layermask_personmaskultra_v2_192, 0),
+        )
+
+        image_comparer_rgthree_486 = image_comparer_rgthree.compare_images(
+            image_a=get_value_at_index(layerutility_imageblend_v2_314, 0),
+            image_b=get_value_at_index(
+                layerutility_imagescalebyaspectratio_v2_331, 0
+            ),
+        )
+
+        saveimage_680 = saveimage.save_images(
+            filename_prefix="ComfyUI",
+            images=get_value_at_index(layerutility_imageblend_v2_314, 0),
+        )
+
+        saved_path = f"output/{saveimage_680['ui']['images'][0]['filename']}"
+
+
+        facesegmentation_838 = facesegmentation.segment(
+            area="face+forehead (if available)",
+            grow=0,
+            grow_tapered=False,
+            blur=13,
+            analysis_models=get_value_at_index(faceanalysismodels_840, 0),
+            image=get_value_at_index(layerutility_imageblend_v2_399, 0),
+        )
+
+        growmask_839 = growmask.expand_mask(
+            expand=0,
+            tapered_corners=True,
+            mask=get_value_at_index(facesegmentation_838, 0),
+        )
+
+        layerutility_imageblend_v2_686 = layerutility_imageblend_v2.image_blend_v2(
+            invert_mask=False,
+            blend_mode="normal",
+            opacity=100,
+            background_image=get_value_at_index(layerutility_imageblend_v2_314, 0),
+            layer_image=get_value_at_index(layerutility_imageblend_v2_399, 0),
+            layer_mask=get_value_at_index(growmask_839, 0),
+        )
+
+        image_comparer_rgthree_820 = image_comparer_rgthree.compare_images(
+            image_a=get_value_at_index(layerutility_imageblend_v2_399, 0),
+            image_b=get_value_at_index(
+                layerutility_imagescalebyaspectratio_v2_331, 0
+            ),
+        )
+
+        imageupscalewithmodel_831 = imageupscalewithmodel.upscale(
+            upscale_model=get_value_at_index(upscalemodelloader_835, 0),
+            image=get_value_at_index(layerutility_imageblend_v2_686, 0),
         )
-        saved_path = f"output/{saved['ui']['images'][0]['filename']}"
+
         return saved_path
 
-# Create Gradio interface
-
-examples = [
-    ["", "mona.png", "receita-tacos.webp", 15, 0.6],
-    ["a woman looking at a house catching fire on the background", "disaster_girl.png", "abaporu.jpg", 15, 0.15],
-    ["istanbul aerial, dramatic photography", "natasha.png", "istambul.jpg", 15, 0.5],
-]
-
-output_image = gr.Image(label="Generated Image")
-
-with gr.Blocks() as app:
-    gr.Markdown("# FLUX Style Shaping")
-    gr.Markdown("Flux[dev] Redux + Flux[dev] Depth ComfyUI workflow by [Nathan Shipley](https://x.com/CitizenPlain) running directly on Gradio. [workflow](https://gist.github.com/nathanshipley/7a9ac1901adde76feebe58d558026f68) - [how to convert your any comfy workflow to gradio](https://huggingface.co/blog/run-comfyui-workflows-on-spaces)")
-    with gr.Row():
-        with gr.Column():
-            prompt_input = gr.Textbox(label="Prompt", placeholder="Enter your prompt here...")
-            with gr.Row():
-                with gr.Group():
-                    structure_image = gr.Image(label="Structure Image", type="filepath")
-                    depth_strength = gr.Slider(minimum=0, maximum=50, value=15, label="Depth Strength")
-                with gr.Group():
-                    style_image = gr.Image(label="Style Image", type="filepath")
-                    style_strength = gr.Slider(minimum=0, maximum=1, value=0.5, label="Style Strength")
+
+if __name__ == "__main__":
+    # main()
+
+    with gr.Blocks() as app:
+        gr.Markdown("# Swap Hairstyle")
+
+        with gr.Row():
+            # 添加输入
+            with gr.Column():
+                with gr.Row():
+                    # 第一组包括结构图像和深度强度
+                    with gr.Group():
+                        model_image = gr.Image(label="Model Image", type="filepath")
+                    # 第二组包括风格图像和风格强度
+                    with gr.Group():
+                        hairstyle_template_image = gr.Image(label="Hairstyle Template Image", type="filepath")
+                
+            with gr.Column():
+                # 输出图像
+                output_image = gr.Image(label="Generated Image")
+        
+        with gr.Row():
             generate_btn = gr.Button("Generate")
-            
-            gr.Examples(
-                examples=examples,
-                inputs=[prompt_input, structure_image, style_image, depth_strength, style_strength],
-                outputs=[output_image],
-                fn=generate_image,
-                cache_examples=True,
-                cache_mode="lazy"
-            )
+
+        generate_btn.click(
+            fn=generate_image,
+            inputs=[model_image, hairstyle_template_image],
+            outputs=[output_image]
+        )
         
-        with gr.Column():
-            output_image.render()
-    generate_btn.click(
-        fn=generate_image,
-        inputs=[prompt_input, structure_image, style_image, depth_strength, style_strength],
-        outputs=[output_image]
-    )
+        app.launch(share=True)
 
-if __name__ == "__main__":
-    app.launch(share=True)

app.py.bak

@@ -0,0 +1,340 @@
+import os
+import random
+import sys
+from typing import Sequence, Mapping, Any, Union
+import torch
+import gradio as gr
+from PIL import Image
+from huggingface_hub import hf_hub_download
+import spaces
+from comfy import model_management
+
+hf_hub_download(repo_id="black-forest-labs/FLUX.1-Redux-dev", filename="flux1-redux-dev.safetensors", local_dir="models/style_models")
+hf_hub_download(repo_id="black-forest-labs/FLUX.1-Depth-dev", filename="flux1-depth-dev.safetensors", local_dir="models/diffusion_models")
+hf_hub_download(repo_id="Comfy-Org/sigclip_vision_384", filename="sigclip_vision_patch14_384.safetensors", local_dir="models/clip_vision")
+hf_hub_download(repo_id="Kijai/DepthAnythingV2-safetensors", filename="depth_anything_v2_vitl_fp32.safetensors", local_dir="models/depthanything")
+hf_hub_download(repo_id="black-forest-labs/FLUX.1-dev", filename="ae.safetensors", local_dir="models/vae/FLUX1")
+hf_hub_download(repo_id="comfyanonymous/flux_text_encoders", filename="clip_l.safetensors", local_dir="models/text_encoders")
+t5_path = hf_hub_download(repo_id="comfyanonymous/flux_text_encoders", filename="t5xxl_fp16.safetensors", local_dir="models/text_encoders/t5")
+
+# Import all the necessary functions from the original script
+def get_value_at_index(obj: Union[Sequence, Mapping], index: int) -> Any:
+    try:
+        return obj[index]
+    except KeyError:
+        return obj["result"][index]
+
+# Add all the necessary setup functions from the original script
+def find_path(name: str, path: str = None) -> str:
+    if path is None:
+        path = os.getcwd()
+    if name in os.listdir(path):
+        path_name = os.path.join(path, name)
+        print(f"{name} found: {path_name}")
+        return path_name
+    parent_directory = os.path.dirname(path)
+    if parent_directory == path:
+        return None
+    return find_path(name, parent_directory)
+
+def add_comfyui_directory_to_sys_path() -> None:
+    comfyui_path = find_path("ComfyUI")
+    if comfyui_path is not None and os.path.isdir(comfyui_path):
+        sys.path.append(comfyui_path)
+        print(f"'{comfyui_path}' added to sys.path")
+
+def add_extra_model_paths() -> None:
+    try:
+        from main import load_extra_path_config
+    except ImportError:
+        from utils.extra_config import load_extra_path_config
+    extra_model_paths = find_path("extra_model_paths.yaml")
+    if extra_model_paths is not None:
+        load_extra_path_config(extra_model_paths)
+    else:
+        print("Could not find the extra_model_paths config file.")
+
+# Initialize paths
+add_comfyui_directory_to_sys_path()
+add_extra_model_paths()
+
+def import_custom_nodes() -> None:
+    import asyncio
+    import execution
+    from nodes import init_extra_nodes
+    import server
+    loop = asyncio.new_event_loop()
+    asyncio.set_event_loop(loop)
+    server_instance = server.PromptServer(loop)
+    execution.PromptQueue(server_instance)
+    init_extra_nodes()
+
+# Import all necessary nodes
+from nodes import (
+    StyleModelLoader,
+    VAEEncode,
+    NODE_CLASS_MAPPINGS,
+    LoadImage,
+    CLIPVisionLoader,
+    SaveImage,
+    VAELoader,
+    CLIPVisionEncode,
+    DualCLIPLoader,
+    EmptyLatentImage,
+    VAEDecode,
+    UNETLoader,
+    CLIPTextEncode,
+)
+
+# Initialize all constant nodes and models in global context
+import_custom_nodes()
+
+# Global variables for preloaded models and constants
+#with torch.inference_mode():
+    # Initialize constants
+intconstant = NODE_CLASS_MAPPINGS["INTConstant"]()
+CONST_1024 = intconstant.get_value(value=1024)
+
+# Load CLIP
+dualcliploader = DualCLIPLoader()
+CLIP_MODEL = dualcliploader.load_clip(
+    clip_name1="t5/t5xxl_fp16.safetensors",
+    clip_name2="clip_l.safetensors",
+    type="flux",
+)
+
+# Load VAE
+vaeloader = VAELoader()
+VAE_MODEL = vaeloader.load_vae(vae_name="FLUX1/ae.safetensors")
+
+# Load UNET
+unetloader = UNETLoader()
+UNET_MODEL = unetloader.load_unet(
+    unet_name="flux1-depth-dev.safetensors", weight_dtype="default"
+)
+
+# Load CLIP Vision
+clipvisionloader = CLIPVisionLoader()
+CLIP_VISION_MODEL = clipvisionloader.load_clip(
+    clip_name="sigclip_vision_patch14_384.safetensors"
+)
+
+# Load Style Model
+stylemodelloader = StyleModelLoader()
+STYLE_MODEL = stylemodelloader.load_style_model(
+    style_model_name="flux1-redux-dev.safetensors"
+)
+
+# Initialize samplers
+ksamplerselect = NODE_CLASS_MAPPINGS["KSamplerSelect"]()
+SAMPLER = ksamplerselect.get_sampler(sampler_name="euler")
+
+# Initialize depth model
+cr_clip_input_switch = NODE_CLASS_MAPPINGS["CR Clip Input Switch"]()
+downloadandloaddepthanythingv2model = NODE_CLASS_MAPPINGS["DownloadAndLoadDepthAnythingV2Model"]()
+DEPTH_MODEL = downloadandloaddepthanythingv2model.loadmodel(
+    model="depth_anything_v2_vitl_fp32.safetensors"
+)
+
+cliptextencode = CLIPTextEncode()
+loadimage = LoadImage()
+vaeencode = VAEEncode()
+fluxguidance = NODE_CLASS_MAPPINGS["FluxGuidance"]()
+instructpixtopixconditioning = NODE_CLASS_MAPPINGS["InstructPixToPixConditioning"]()
+clipvisionencode = CLIPVisionEncode()
+stylemodelapplyadvanced = NODE_CLASS_MAPPINGS["StyleModelApplyAdvanced"]()
+emptylatentimage = EmptyLatentImage()
+basicguider = NODE_CLASS_MAPPINGS["BasicGuider"]()
+basicscheduler = NODE_CLASS_MAPPINGS["BasicScheduler"]()        
+randomnoise = NODE_CLASS_MAPPINGS["RandomNoise"]()
+samplercustomadvanced = NODE_CLASS_MAPPINGS["SamplerCustomAdvanced"]()
+vaedecode = VAEDecode()
+cr_text = NODE_CLASS_MAPPINGS["CR Text"]()
+saveimage = SaveImage()
+getimagesizeandcount = NODE_CLASS_MAPPINGS["GetImageSizeAndCount"]()
+depthanything_v2 = NODE_CLASS_MAPPINGS["DepthAnything_V2"]()
+imageresize = NODE_CLASS_MAPPINGS["ImageResize+"]()
+
+model_loaders = [CLIP_MODEL, VAE_MODEL, UNET_MODEL, CLIP_VISION_MODEL]
+
+model_management.load_models_gpu([
+    loader[0].patcher if hasattr(loader[0], 'patcher') else loader[0] for loader in model_loaders
+])
+
+@spaces.GPU
+def generate_image(prompt, structure_image, style_image, depth_strength=15, style_strength=0.5) -> str:
+    """Main generation function that processes inputs and returns the path to the generated image."""
+    with torch.inference_mode():
+        # Set up CLIP
+        clip_switch = cr_clip_input_switch.switch(
+            Input=1,
+            clip1=get_value_at_index(CLIP_MODEL, 0),
+            clip2=get_value_at_index(CLIP_MODEL, 0),
+        )
+        
+        # Encode text
+        text_encoded = cliptextencode.encode(
+            text=prompt,
+            clip=get_value_at_index(clip_switch, 0),
+        )
+        empty_text = cliptextencode.encode(
+            text="",
+            clip=get_value_at_index(clip_switch, 0),
+        )
+        
+        # Process structure image
+        structure_img = loadimage.load_image(image=structure_image)
+        
+        # Resize image
+        resized_img = imageresize.execute(
+            width=get_value_at_index(CONST_1024, 0),
+            height=get_value_at_index(CONST_1024, 0),
+            interpolation="bicubic",
+            method="keep proportion",
+            condition="always",
+            multiple_of=16,
+            image=get_value_at_index(structure_img, 0),
+        )
+        
+        # Get image size
+        size_info = getimagesizeandcount.getsize(
+            image=get_value_at_index(resized_img, 0)
+        )
+        
+        # Encode VAE
+        vae_encoded = vaeencode.encode(
+            pixels=get_value_at_index(size_info, 0),
+            vae=get_value_at_index(VAE_MODEL, 0),
+        )
+        
+        # Process depth
+        depth_processed = depthanything_v2.process(
+            da_model=get_value_at_index(DEPTH_MODEL, 0),
+            images=get_value_at_index(size_info, 0),
+        )
+        
+        # Apply Flux guidance
+        flux_guided = fluxguidance.append(
+            guidance=depth_strength,
+            conditioning=get_value_at_index(text_encoded, 0),
+        )
+        
+        # Process style image
+        style_img = loadimage.load_image(image=style_image)
+        
+        # Encode style with CLIP Vision
+        style_encoded = clipvisionencode.encode(
+            crop="center",
+            clip_vision=get_value_at_index(CLIP_VISION_MODEL, 0),
+            image=get_value_at_index(style_img, 0),
+        )
+        
+        # Set up conditioning
+        conditioning = instructpixtopixconditioning.encode(
+            positive=get_value_at_index(flux_guided, 0),
+            negative=get_value_at_index(empty_text, 0),
+            vae=get_value_at_index(VAE_MODEL, 0),
+            pixels=get_value_at_index(depth_processed, 0),
+        )
+        
+        # Apply style
+        style_applied = stylemodelapplyadvanced.apply_stylemodel(
+            strength=style_strength,
+            conditioning=get_value_at_index(conditioning, 0),
+            style_model=get_value_at_index(STYLE_MODEL, 0),
+            clip_vision_output=get_value_at_index(style_encoded, 0),
+        )
+        
+        # Set up empty latent
+        empty_latent = emptylatentimage.generate(
+            width=get_value_at_index(resized_img, 1),
+            height=get_value_at_index(resized_img, 2),
+            batch_size=1,
+        )
+        
+        # Set up guidance
+        guided = basicguider.get_guider(
+            model=get_value_at_index(UNET_MODEL, 0),
+            conditioning=get_value_at_index(style_applied, 0),
+        )
+        
+        # Set up scheduler
+        schedule = basicscheduler.get_sigmas(
+            scheduler="simple",
+            steps=28,
+            denoise=1,
+            model=get_value_at_index(UNET_MODEL, 0),
+        )
+        
+        # Generate random noise
+        noise = randomnoise.get_noise(noise_seed=random.randint(1, 2**64))
+        
+        # Sample
+        sampled = samplercustomadvanced.sample(
+            noise=get_value_at_index(noise, 0),
+            guider=get_value_at_index(guided, 0),
+            sampler=get_value_at_index(SAMPLER, 0),
+            sigmas=get_value_at_index(schedule, 0),
+            latent_image=get_value_at_index(empty_latent, 0),
+        )
+        
+        # Decode VAE
+        decoded = vaedecode.decode(
+            samples=get_value_at_index(sampled, 0),
+            vae=get_value_at_index(VAE_MODEL, 0),
+        )
+        
+        # Save image
+        prefix = cr_text.text_multiline(text="Flux_BFL_Depth_Redux")
+        
+        saved = saveimage.save_images(
+            filename_prefix=get_value_at_index(prefix, 0),
+            images=get_value_at_index(decoded, 0),
+        )
+        saved_path = f"output/{saved['ui']['images'][0]['filename']}"
+        return saved_path
+
+# Create Gradio interface
+
+examples = [
+    ["", "mona.png", "receita-tacos.webp", 15, 0.6],
+    ["a woman looking at a house catching fire on the background", "disaster_girl.png", "abaporu.jpg", 15, 0.15],
+    ["istanbul aerial, dramatic photography", "natasha.png", "istambul.jpg", 15, 0.5],
+]
+
+output_image = gr.Image(label="Generated Image")
+
+with gr.Blocks() as app:
+    gr.Markdown("# FLUX Style Shaping")
+    gr.Markdown("Flux[dev] Redux + Flux[dev] Depth ComfyUI workflow by [Nathan Shipley](https://x.com/CitizenPlain) running directly on Gradio. [workflow](https://gist.github.com/nathanshipley/7a9ac1901adde76feebe58d558026f68) - [how to convert your any comfy workflow to gradio](https://huggingface.co/blog/run-comfyui-workflows-on-spaces)")
+    with gr.Row():
+        with gr.Column():
+            prompt_input = gr.Textbox(label="Prompt", placeholder="Enter your prompt here...")
+            with gr.Row():
+                with gr.Group():
+                    structure_image = gr.Image(label="Structure Image", type="filepath")
+                    depth_strength = gr.Slider(minimum=0, maximum=50, value=15, label="Depth Strength")
+                with gr.Group():
+                    style_image = gr.Image(label="Style Image", type="filepath")
+                    style_strength = gr.Slider(minimum=0, maximum=1, value=0.5, label="Style Strength")
+            generate_btn = gr.Button("Generate")
+            
+            gr.Examples(
+                examples=examples,
+                inputs=[prompt_input, structure_image, style_image, depth_strength, style_strength],
+                outputs=[output_image],
+                fn=generate_image,
+                cache_examples=True,
+                cache_mode="lazy"
+            )
+        
+        with gr.Column():
+            output_image.render()
+    generate_btn.click(
+        fn=generate_image,
+        inputs=[prompt_input, structure_image, style_image, depth_strength, style_strength],
+        outputs=[output_image]
+    )
+
+if __name__ == "__main__":
+    app.launch(share=True)