Update app_v4.py

app_v4.py

@@ -150,6 +150,8 @@ def generate_focus(control_image, focus_list):
         print(f"Error generating focus: {e}")
         return "highly detailed photo, raw photography.", "Original Image Dimensions: N/A"
         
+
+
 @spaces.GPU(duration=6, progress=gr.Progress(track_tqdm=True))
 @torch.no_grad()
 def generate_image(prompt, scale, steps, control_image, controlnet_conditioning_scale, guidance_scale, seed, guidance_end):
@@ -162,21 +164,24 @@ def generate_image(prompt, scale, steps, control_image, controlnet_conditioning_
     # Crop to nearest multiple of 32
     w = w - w % 32
     h = h - h % 32
+    # Fix the resizing code (assuming scale is defined)
     control_image = control_image.resize((w, h), resample=2)
     print("Size to: " + str(control_image.size[0]) + ", " + str(control_image.size[1]))
     print(f"PromptLog: {repr(prompt)}")
     # Convert image to RGB for processing, but keep alpha channel for transparency
     control_image_rgb = control_image.convert("RGB")
     control_image_alpha = control_image.split()[-1]
-    # Convert alpha channel to a mask using numpy array (0 for transparent, 1 for opaque)
+    # Convert alpha channel to a mask (0 for transparent, 1 for opaque)
     alpha = np.array(control_image_alpha)
     mask = (alpha > 128).astype(np.float32)  # Create binary mask with 0 and 1
+    # Convert control_image_rgb to numpy array (if necessary)
+    control_image_rgb_np = np.array(control_image_rgb)
     with torch.inference_mode():
         image = pipe(
             generator=generator,
             prompt=prompt,
-            control_image=control_image_rgb,
-            mask_image=mask,  # Pass the numpy array as mask_image
+            control_image=control_image_rgb_np,  # Use numpy array
+            mask_image=mask,  # Pass numpy mask
             controlnet_conditioning_scale=controlnet_conditioning_scale,
             num_inference_steps=steps,
             guidance_scale=guidance_scale,