Update app.py

app.py

@@ -110,7 +110,10 @@ def process_video(radio, video, target_language):
 
     video_path_fix = video_path
 
-    has_face = check_for_faces(video_path)
+    if has_closeup_face:
+        has_face = True
+    else:
+        has_face = check_for_faces(video_path)
 
     if has_face:
         cmd = f"python Wav2Lip/inference.py --checkpoint_path 'Wav2Lip/checkpoints/wav2lip_gan.pth' --face {shlex.quote(video_path)} --audio '{run_uuid}_output_synth.wav' --pads {pad_top} {pad_bottom} {pad_left} {pad_right} --resize_factor {rescaleFactor} --nosmooth --outfile '{run_uuid}_output_video.mp4'"
@@ -155,6 +158,7 @@ iface = gr.Interface(
         radio,
         video,
         gr.Dropdown(choices=["English", "Spanish", "French", "German", "Italian", "Portuguese", "Polish", "Turkish", "Russian", "Dutch", "Czech", "Arabic", "Chinese (Simplified)"], label="Target Language for Dubbing", value="Spanish")
+        checkbox = gr.Checkbox(label="Video has a close-up face", default=False)    
     ],
     outputs=gr.Video(),
     live=False,
@@ -172,6 +176,7 @@ with gr.Blocks() as demo:
     - The tool uses open-source models for all models. It's a alpha version.
     - Quality can be improved but would require more processing time per video. For scalability and hardware limitations, speed was chosen, not just quality.
     - If you need more than 1 minute, duplicate the Space and change the limit on app.py.
+    - If you incorrectly mark the 'Video has a close-up face' checkbox, the dubbing may not work as expected.
     """)
 demo.queue(concurrency_count=1, max_size=15)
 demo.launch()