diff --git a/README.md b/README.md
index a278f7a..5920182 100644
--- a/README.md
+++ b/README.md
@@ -17,16 +17,20 @@ def prompt_worker(q, server):
 
     # This is a workaround since using blender as a python module, the bpy module has to be imported after in the custom thread, otherwise it will cause a segfault if imported in the custom nodes.
 
-    # Add this line here
-    import sys
-    sys.path.append(os.path.join(os.path.dirname(__file__), "custom_nodes", "avatar-graph-comfyui"))
+    # Add next line
     import global_bpy
 
     while True:
         item, item_id = q.get()
-        # Add this line here
+        # Add next line
         global_bpy.reset_bpy()
 
+if __name__ == "__main__":
+    # Add next two lines 
+    import sys
+    sys.path.append(os.path.join(os.path.dirname(__file__), "custom_nodes", "avatar-graph-comfyui"))
+
+
 ```
 
 5. Restart comfyui
diff --git a/__init__.py b/__init__.py
index 51b16ca..e0dd9ae 100644
--- a/__init__.py
+++ b/__init__.py
@@ -4,6 +4,7 @@
 @nickname: Avatar Graph
 @description: Include nodes for sam + bpy operation, that allows workflow creations for generative 2d character rig.
 """
+import routes
 
 import inspect
 import os
diff --git a/routes.py b/routes.py
new file mode 100644
index 0000000..fa85b75
--- /dev/null
+++ b/routes.py
@@ -0,0 +1,85 @@
+from aiohttp import web
+from segment_anything import sam_model_registry, SamPredictor
+from PIL import Image, ImageOps
+import os
+import requests
+import folder_paths
+import json
+import numpy as np
+import server
+
+# For speeding up ONNX model, see https://github.com/facebookresearch/segment-anything/tree/main/demo#onnx-multithreading-with-sharedarraybuffer
+def inject_headers(original_handler):
+    async def _handler(request):
+        res = await original_handler(request)
+        res.headers["Cross-Origin-Opener-Policy"] = "same-origin"
+        res.headers["Cross-Origin-Embedder-Policy"] = "credentialless"
+        return res
+
+    return _handler
+
+
+routes = []
+for item in server.PromptServer.instance.routes._items:
+    if item.path == "/":
+        item = web.RouteDef(
+            method=item.method,
+            path=item.path,
+            handler=inject_headers(item.handler),
+            kwargs=item.kwargs,
+        )
+    routes.append(item)
+server.PromptServer.instance.routes._items = routes
+
+
+@server.PromptServer.instance.routes.get("/sam_model")
+async def get_sam_model(request):
+    filename = os.path.join(folder_paths.base_path, "web/models/sam.onnx")
+    # print(filename)
+    if not os.path.isfile(filename):
+        os.makedirs(os.path.dirname(filename), exist_ok=True)
+        response = requests.get(
+            "https://avatech-avatar-dev1.nyc3.cdn.digitaloceanspaces.com/models/sam.onnx"
+        )
+        response.raise_for_status()
+        with open(filename, "wb") as f:
+            f.write(response.content)
+        print(f"ONNX model downloaded: {filename}")
+    return web.FileResponse(filename)
+
+
+def load_image(image):
+    image_path = folder_paths.get_annotated_filepath(image)
+    i = Image.open(image_path)
+    i = ImageOps.exif_transpose(i)
+    image = i.convert("RGB")
+    image = np.array(image).astype(np.float32) / 255.0
+    return image
+
+
+@server.PromptServer.instance.routes.post("/sam_model")
+async def post_sam_model(request):
+    post = await request.json()
+    emb_id = post.get("embedding_id")
+    emb_filename = f"{folder_paths.get_output_directory()}/{emb_id}.npy"
+    if not os.path.exists(emb_filename):
+        image = load_image(post.get("image"))
+        ckpt = post.get("ckpt")
+        model_type = post.get("model_type")
+        ckpt = folder_paths.get_full_path("sam", ckpt)
+        sam = sam_model_registry[model_type](checkpoint=ckpt)
+        predictor = SamPredictor(sam)
+
+        image_np = (image * 255).astype(np.uint8)
+        predictor.set_image(image_np)
+        emb = predictor.get_image_embedding().cpu().numpy()
+        np.save(emb_filename, emb)
+        with open(f"{folder_paths.get_output_directory()}/{emb_id}.json", "w") as f:
+            json.dump(
+                {
+                    "input_size": predictor.input_size,
+                    "original_size": predictor.original_size,
+                },
+                f,
+            )
+    return web.json_response({})