Update app.py

app.py

@@ -111,15 +111,17 @@ def load_control_lora(control_type, current_lora_state):
         print(f"Error loading {control_type} LoRA: {e}")
         raise
 
-def process_video_for_canny(video):
+def process_video_for_canny(video, width, height):
     """
     Process video for canny control.
     """
     print("Processing video for canny control...")
     canny_video = []
+    detect_resolution = video[0].size()
+    
     for frame in video: 
         # TODO: change resolution logic
-        canny_video.append(canny_processor(frame, low_threshold=50, high_threshold=200, detect_resolution=1024, image_resolution=1024))
+        canny_video.append(canny_processor(frame, low_threshold=50, high_threshold=200, detect_resolution=detect_resolution, image_resolution=(width, height)))
      
     return canny_video
 
@@ -169,7 +171,7 @@ def process_video_for_pose(video):
     return pose_video
 
 @spaces.GPU()
-def process_input_video(reference_video):
+def process_input_video(reference_video, width, height):
     """
     Process the input video for canny edges and return both processed video and preview.
     """
@@ -181,7 +183,7 @@ def process_input_video(reference_video):
         video = load_video(reference_video)
         
         # Process video for canny edges
-        processed_video = process_video_for_canny(video)
+        processed_video = process_video_for_canny(video, width, height)
         
         # Create a preview video file for display
         fps = 24
@@ -195,13 +197,13 @@ def process_input_video(reference_video):
         print(f"Error processing input video: {e}")
         return None
 
-def process_video_for_control(reference_video, control_type):
+def process_video_for_control(reference_video, control_type, width, height):
     """Process video based on the selected control type - now only used for non-canny types"""
     video = load_video(reference_video)
     
     if control_type == "canny":
         # This should not be called for canny since it's pre-processed
-        processed_video = process_video_for_canny(video)
+        processed_video = process_video_for_canny(video, width, height)
     elif control_type == "depth":
         processed_video = process_video_for_depth(video)
     elif control_type == "pose":
@@ -256,7 +258,7 @@ def generate_video(
             processed_video = load_video(control_video)
         else:
             # Fallback to processing on-demand for other control types
-            processed_video = process_video_for_control(reference_video, control_type)
+            processed_video = process_video_for_control(reference_video, control_type, width, height)
         
         # Convert to tensor
         processed_video = read_video(processed_video)
@@ -345,8 +347,7 @@ with gr.Blocks() as demo:
     )
     
     # State variables
-    current_lora_state = gr.State(value=None)
-    processed_video_state = gr.State(value=None)  # Store processed video frames
+    #current_lora_state = gr.State(value=None)
     
     with gr.Row():
         with gr.Column(scale=1):
@@ -475,7 +476,7 @@ with gr.Blocks() as demo:
     # Auto-process video when uploaded
     reference_video.upload(
         fn=process_input_video,
-        inputs=[reference_video],
+        inputs=[reference_video, width, height],
         outputs=[control_video],
         show_progress=True
     )