Update app.py

app.py

@@ -59,6 +59,7 @@ canny_processor = CannyDetector()
 
 current_dir = Path(__file__).parent
 
+
 def cleanup_session_files(request: gr.Request):
     """Clean up session-specific temporary files when user disconnects"""
     try:
@@ -133,7 +134,8 @@ def process_video_for_canny(video, width, height):
     """
     print("Processing video for canny control...")
     canny_video = []
-    detect_resolution = max(video[0].size[0],video[0].size[1])
+    print(video[0].shape[0], video[0].shape[1])
+    detect_resolution = max(video[0].shape[0],video[0].shape[1])
     image_resolution = max(width, height)
     for frame in video: 
         canny_video.append(canny_processor(frame, low_threshold=50, high_threshold=200, detect_resolution=detect_resolution, image_resolution=image_resolution, output_type="np"))
@@ -145,12 +147,19 @@ def process_input_video(reference_video, width, height,
     """
     Process the input video for canny edges and return both processed video and preview.
     """
+    print(current_dir)
     if reference_video is None:
         return None
     
     try:
         # Load video into a list of PIL images
-        video = load_video(reference_video)
+        #video = load_video(reference_video)
+        video = []
+        with imageio.get_reader(reference_video) as reader:
+            # Read all frames
+            for frame in reader:
+                video.append(frame)
+        #video is a numpy array
         
         # Process video for canny edges
         processed_video = process_video_for_canny(video, width, height)
@@ -170,7 +179,13 @@ def process_input_video(reference_video, width, height,
 
 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)
+    #video = load_video(reference_video)
+
+    video = []
+    with imageio.get_reader(reference_video) as reader:
+        # Read all frames
+        for frame in reader:
+            video.append(frame)
     
     if control_type == "canny":
         # This should not be called for canny since it's pre-processed
@@ -232,6 +247,16 @@ def generate_video(
         # Convert to tensor
         processed_video = read_video(processed_video)
         print(type(processed_video))
+
+        # Clean up example video if it's cached in tmp
+        if control_video and "/tmp" in control_video and any(x in control_video for x in ["vid_1_canny", "vid_2_canny", "vid_3_canny", "vid_4_canny"]):
+            if os.path.exists(control_video):
+                try:
+                    os.remove(control_video)
+                    print(f"✓ Deleted cached example video: {control_video}")
+                except OSError as e:
+                    print(f"✗ Failed to delete {control_video}: {e}")
+                
         
         
         # Calculate downscaled dimensions
@@ -305,7 +330,7 @@ def generate_video(
         return None, seed
 
 # Create Gradio interface
-with gr.Blocks(theme=gr.themes.Ocean(font=[gr.themes.GoogleFont("Lexend Deca"), "sans-serif"]), delete_cache=(60, 3600)) as demo:
+with gr.Blocks(theme=gr.themes.Ocean(font=[gr.themes.GoogleFont("Lexend Deca"), "sans-serif"]), delete_cache=(60, 1800)) as demo:
     gr.Markdown(
         """
         # Canny Control LTX Video Distilled