Update app.py

app.py

@@ -9,6 +9,7 @@ from transformers import T5EncoderModel, T5Tokenizer
 from datetime import datetime
 import random
 from moviepy.editor import VideoFileClip
+import ffmpeg
 
 from huggingface_hub import hf_hub_download
 
@@ -59,9 +60,11 @@ def infer(image_path, prompt, orbit_type, progress=gr.Progress(track_tqdm=True))
     # Load and get original image dimensions
     image = load_image(image_path)
     original_width, original_height = image.size
+    print(f"IMAGE INPUT SIZE: {original_width} x {original_height}")
     
     # Calculate target dimensions maintaining aspect ratio
     target_width, target_height = calculate_resize_dimensions(original_width, original_height)
+    print(f"TARGET SIZE: {target_width} x {target_height}")
     
     lora_path = "checkpoints/"
     weight_name = "orbit_left_lora_weights.safetensors" if orbit_type == "Left" else "orbit_up_lora_weights.safetensors"
@@ -105,25 +108,21 @@ def infer(image_path, prompt, orbit_type, progress=gr.Progress(track_tqdm=True))
     # First export the original video
     export_to_video(video.frames[0], temp_path, fps=8)
     
-    # Then resize it with moviepy
     try:
-        video_clip = VideoFileClip(temp_path)
-        resized_clip = video_clip.resize(width=target_width, height=target_height)
-        resized_clip.write_videofile(
-            final_path,
-            codec='libx264',
-            fps=8,
-            preset='medium',
-            ffmpeg_params=['-crf', '23'],
-            verbose=False,
-            logger=None
-        )
+        # Use ffmpeg-python
+        stream = ffmpeg.input(temp_path)
+        stream = ffmpeg.filter(stream, 'scale', target_width, target_height)
+        stream = ffmpeg.output(stream, final_path, 
+                             vcodec='libx264', 
+                             preset='medium', 
+                             crf=23, 
+                             acodec='copy')
+        ffmpeg.run(stream, overwrite_output=True, capture_stdout=True, capture_stderr=True)
+    except ffmpeg.Error as e:
+        print('stdout:', e.stdout.decode('utf8'))
+        print('stderr:', e.stderr.decode('utf8'))
+        raise e
     finally:
-        # Make sure we clean up the clips
-        if 'video_clip' in locals():
-            video_clip.close()
-        if 'resized_clip' in locals():
-            resized_clip.close()
         if os.path.exists(temp_path):
             os.remove(temp_path)