Update app.py

app.py

@@ -100,31 +100,6 @@ class RetroArtConverter:
         try:
             self.zoe_depth = ZoeDetector.from_pretrained("lllyasviel/Annotators")
             self.zoe_depth.to(self.device)
-
-            # --- MONKEY-PATCH FIX for numpy.int64 TypeError ---
-            # The ZoeDetector's internal Midas model (self.zoe_depth.model.core.prep.resizer)
-            # has its 'height' and 'width' attributes as numpy.int64,
-            # which torch.nn.functional.interpolate dislikes.
-            # We explicitly cast them to standard Python int.
-            if hasattr(self.zoe_depth, 'model') and \
-               hasattr(self.zoe_depth.model, 'core') and \
-               hasattr(self.zoe_depth.model.core, 'prep') and \
-               hasattr(self.zoe_depth.model.core.prep, 'resizer'):
-                
-                print("Applying monkey-patch to ZoeDepth model resizer...")
-                resizer = self.zoe_depth.model.core.prep.resizer
-                
-                if hasattr(resizer, 'height') and not isinstance(resizer.height, int):
-                    resizer.height = int(resizer.height)
-                    print(f"  Patched resizer.height to type: {type(resizer.height)}")
-                    
-                if hasattr(resizer, 'width') and not isinstance(resizer.width, int):
-                    resizer.width = int(resizer.width)
-                    print(f"  Patched resizer.width to type: {type(resizer.width)}")
-                    
-                print("Monkey-patch applied.")
-            # --- END MONKEY-PATCH ---
-
             print("✓ Zoe Depth loaded successfully")
             self.models_loaded['zoe_depth'] = True
         except Exception as e:
@@ -262,20 +237,8 @@ class RetroArtConverter:
     def get_depth_map(self, image):
         """Generate depth map using Zoe Depth"""
         if self.zoe_depth is not None:
-            # Get the size from the PIL image
-            w, h = image.size 
-            
-            # Ensure they are standard Python ints, not numpy.int64
-            w_int, h_int = int(w), int(h)
-            
-            # detect_resolution expects an INT (for shortest side)
-            # image_resolution expects a TUPLE (width, height)
-            print(f"  Generating Zoe depth map at {w_int}x{h_int} resolution...")
-            depth_image = self.zoe_depth(
-                image, 
-                detect_resolution=512,  # Use a standard int value
-                image_resolution=(w_int, h_int) # Pass the tuple here
-            )
+            # Use Zoe detector
+            depth_image = self.zoe_depth(image, detect_resolution=512, image_resolution=1024)
             return depth_image
         else:
             # Fallback to simple grayscale
@@ -341,24 +304,18 @@ class RetroArtConverter:
         original_width, original_height = input_image.size
         target_width, target_height = self.calculate_optimal_size(original_width, original_height)
         
-        # Cast to int() to prevent numpy.int64 errors
-        target_width = int(target_width)
-        target_height = int(target_height)
-        
         print(f"Resizing from {original_width}x{original_height} to {target_width}x{target_height}")
         print(f"Prompt: {prompt}")
         print(f"Img2Img Strength: {strength}")
-
-        # 1. Create resized_image FIRST...
-        resized_image = input_image.resize((target_width, target_height), Image.LANCZOS)
         
-        # 2. ...THEN pass it to get_depth_map.
+        # Resize with high quality - ensure dimensions are Python ints
+        resized_image = input_image.resize((int(target_width), int(target_height)), Image.LANCZOS)
+        
+        # Generate depth map using Zoe
         print("Generating Zoe depth map...")
         depth_image = self.get_depth_map(resized_image)
-        
         if depth_image.size != (target_width, target_height):
-            print(f"Warning: Depth map size {depth_image.size} does not match target {target_width}x{target_height}. Resizing...")
-            depth_image = depth_image.resize((target_width, target_height), Image.LANCZOS)
+            depth_image = depth_image.resize((int(target_width), int(target_height)), Image.LANCZOS)
         
         # Handle face detection for InstantID
         using_multiple_controlnets = self.using_multiple_controlnets