self.ip_ckpt

handler.py

@@ -9,6 +9,7 @@ from pathlib import Path
 from huggingface_hub import hf_hub_download, snapshot_download
 from ip_adapter.ip_adapter import IPAdapterXL
 from safetensors.torch import load_file
+import os
 
 from diffusers import (
     ControlNetModel,
@@ -24,8 +25,8 @@ dtype = torch.float16 if str(device).__contains__("cuda") else torch.float32
 
 # initialization
 base_model_path = "stabilityai/stable-diffusion-xl-base-1.0"
-image_encoder_path = "sdxl_models/image_encoder"
-ip_ckpt = "sdxl_models/ip-adapter_sdxl.bin"
+# image_encoder_path = "sdxl_models/image_encoder"
+# ip_ckpt = "sdxl_models/ip-adapter_sdxl.bin"
 controlnet_path = "diffusers/controlnet-canny-sdxl-1.0"
 
 
@@ -33,6 +34,13 @@ controlnet_path = "diffusers/controlnet-canny-sdxl-1.0"
 class EndpointHandler():
     def __init__(self, model_dir):
         
+        repo_id = "h94/IP-Adapter"
+
+        # Descargar todo el contenido del directorio image_encoder
+        local_repo_path = snapshot_download(repo_id=repo_id)
+        # image_encoder_local_path = os.path.join(local_repo_path, "image_encoder")
+        self.image_encoder_local_path = os.path.join(local_repo_path, "sdxl_models", "image_encoder")
+        self.ip_ckpt = os.path.join(local_repo_path, "sdxl_models", "ip-adapter_sdxl.bin")
 
         
         self.controlnet = ControlNetModel.from_pretrained(
@@ -63,8 +71,8 @@ class EndpointHandler():
 
         self.ip_model = IPAdapterXL(
             self.pipe,
-            image_encoder_path,
-            ip_ckpt,
+            self.image_encoder_local_path,
+            self.ip_ckpt,
             device,
             target_blocks=["up_blocks.0.attentions.1"],
         )
@@ -89,14 +97,14 @@ class EndpointHandler():
             if target == "Load original IP-Adapter":
                 # target_blocks=["blocks"] for original IP-Adapter
                 ip_model = IPAdapterXL(
-                    self.pipe, image_encoder_path, ip_ckpt, device, target_blocks=["blocks"]
+                    self.pipe, self.image_encoder_local_path, self.ip_ckpt, device, target_blocks=["blocks"]
                 )
             elif target == "Load only style blocks":
                 # target_blocks=["up_blocks.0.attentions.1"] for style blocks only
                 ip_model = IPAdapterXL(
                     self.pipe,
-                    image_encoder_path,
-                    ip_ckpt,
+                    self.image_encoder_local_path,
+                    self.ip_ckpt,
                     device,
                     target_blocks=["up_blocks.0.attentions.1"],
                 )
@@ -104,8 +112,8 @@ class EndpointHandler():
                 # target_blocks = ["up_blocks.0.attentions.1", "down_blocks.2.attentions.1"] # for style+layout blocks
                 ip_model = IPAdapterXL(
                     self.pipe,
-                    image_encoder_path,
-                    ip_ckpt,
+                    self.image_encoder_local_path,
+                    self.ip_ckpt,
                     device,
                     target_blocks=["up_blocks.0.attentions.1", "down_blocks.2.attentions.1"],
                 )

ip_adapter/ip_adapter.py

@@ -77,22 +77,22 @@ class IPAdapter:
 
         # load image encoder
 
-        repo_id = "h94/IP-Adapter"
+        # repo_id = "h94/IP-Adapter"
 
-        # Descargar todo el contenido del directorio image_encoder
-        local_repo_path = snapshot_download(repo_id=repo_id)
-        # image_encoder_local_path = os.path.join(local_repo_path, "image_encoder")
-        image_encoder_local_path = os.path.join(local_repo_path, "sdxl_models", "image_encoder")
+        # # Descargar todo el contenido del directorio image_encoder
+        # local_repo_path = snapshot_download(repo_id=repo_id)
+        # # image_encoder_local_path = os.path.join(local_repo_path, "image_encoder")
+        # image_encoder_local_path = os.path.join(local_repo_path, "sdxl_models", "image_encoder")
 
-        print("Contenido del directorio image_encoder:", os.listdir(image_encoder_local_path))
+        # print("Contenido del directorio image_encoder:", os.listdir(image_encoder_local_path))
 
-        if "config.json" in os.listdir(image_encoder_local_path) and "pytorch_model.bin" in os.listdir(image_encoder_local_path):
-            # Usar el path local para cargar el modelo
-            self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(
-                image_encoder_local_path
-            ).to(self.device, dtype=torch.float16)
-        else:
-            raise Exception("Los archivos necesarios para cargar el modelo no están en el directorio.")
+        # if "config.json" in os.listdir(image_encoder_local_path) and "pytorch_model.bin" in os.listdir(image_encoder_local_path):
+        #     # Usar el path local para cargar el modelo
+        #     self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(
+        #         image_encoder_local_path
+        #     ).to(self.device, dtype=torch.float16)
+        # else:
+        #     raise Exception("Los archivos necesarios para cargar el modelo no están en el directorio.")
 
 
         # Usar el path local descargado para cargar el modelo
@@ -101,9 +101,9 @@ class IPAdapter:
         # ).to(self.device, dtype=torch.float16)
         
         # load image encoder
-        # self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(self.image_encoder_path).to(
-        #     self.device, dtype=torch.float16
-        # )
+        self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(self.image_encoder_path).to(
+            self.device, dtype=torch.float16
+        )
         
         self.clip_image_processor = CLIPImageProcessor()
         # image proj model