Update app.py

app.py

@@ -206,8 +206,8 @@ def interpolate_video_frames(
     output_fps=hardcoded_fps,
     desired_duration=hardcoded_duration_sec,
     original_duration=hardcoded_duration_sec,
-    output_width=1024,
-    output_height=576,
+    output_width=None,
+    output_height=None,
     use_cuda=False, # this requires FFmpeg to have been compiled with CUDA support (to try - I'm not sure the Hugging Face image has that by default)
     verbose=False):
         
@@ -216,6 +216,10 @@ def interpolate_video_frames(
     filters = []
 
     # Scaling if dimensions are provided
+    # note: upscaling produces disastrous results,
+    # it will double the compute time
+    # I think that's either because we are not hardware-accelerated,
+    # or because of the interpolation done after it, which thus become more computationally intensive
     if output_width and output_height:
         filters.append(f'scale={output_width}:{output_height}')
 
@@ -244,6 +248,7 @@ def interpolate_video_frames(
         '-i', input_file_path,
     ]
 
+    # not supported by the current image, we will have to build it
     if use_cuda:
         cmd.extend(['-hwaccel', 'cuda', '-hwaccel_output_format', 'cuda'])
                
@@ -387,7 +392,7 @@ with gr.Blocks() as demo:
                 minimum=128,
                 maximum=2048,
                 step=32,
-                value=256,
+                value=288,
             )
             select_motion = gr.Dropdown(
                 label='Motion',