Update app.py

app.py

@@ -252,16 +252,35 @@ def generate_motion(source_image_path, driving_audio_path, emotion_name,
     print(f"  Target FPS: {target_fps}")
 
     try:
-        # Call the pipeline's inference method with the WAV audio
-        result_video_path = pipeline.driven_sample(
-            image_path=source_image_path,
-            audio_path=wav_audio_path,
-            cfg_scale=float(cfg_scale),
-            emo=emotion_id,
-            save_dir=".",
-            smooth=smooth_enabled,  # Use the checkbox value
-            silent_audio_path=DEFAULT_SILENT_AUDIO_PATH,
-        )
+        # Temporarily disable smoothing if it causes CUDA tensor issues
+        # Check if smooth causes issues and handle gracefully
+        try:
+            # Try with smoothing first
+            result_video_path = pipeline.driven_sample(
+                image_path=source_image_path,
+                audio_path=wav_audio_path,
+                cfg_scale=float(cfg_scale),
+                emo=emotion_id,
+                save_dir=".",
+                smooth=smooth_enabled,  # Use the checkbox value
+                silent_audio_path=DEFAULT_SILENT_AUDIO_PATH,
+            )
+        except TypeError as tensor_error:
+            if "can't convert cuda" in str(tensor_error) and smooth_enabled:
+                print("Warning: Smoothing caused CUDA tensor error. Retrying without smoothing...")
+                # Retry without smoothing
+                result_video_path = pipeline.driven_sample(
+                    image_path=source_image_path,
+                    audio_path=wav_audio_path,
+                    cfg_scale=float(cfg_scale),
+                    emo=emotion_id,
+                    save_dir=".",
+                    smooth=False,  # Disable smoothing as fallback
+                    silent_audio_path=DEFAULT_SILENT_AUDIO_PATH,
+                )
+                print("Generated video without smoothing due to technical limitations.")
+            else:
+                raise tensor_error
         
         # Apply frame interpolation if requested
         if target_fps > 24:  # Assuming default is around 24 FPS
@@ -351,9 +370,9 @@ with gr.Blocks(theme=gr.themes.Soft(), css=".gradio-container {max-width: 960px
             
             with gr.Row():
                 smooth_checkbox = gr.Checkbox(
-                    label="Enable Smoothing",
-                    value=True,
-                    info="Enables frame smoothing for more natural motion (increases processing time)"
+                    label="Enable Smoothing (Experimental)",
+                    value=False,  # Changed to False due to CUDA issues
+                    info="May cause errors on some systems. If errors occur, disable this option."
                 )
             
             with gr.Row():
@@ -406,9 +425,9 @@ with gr.Blocks(theme=gr.themes.Soft(), css=".gradio-container {max-width: 960px
     # Examples section
     gr.Examples(
         examples=[
-            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "None", 1.0, True, 30],
-            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "Happy", 0.8, True, 30],
-            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "Sad", 1.2, True, 24],
+            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "None", 1.0, False, 30],
+            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "Happy", 0.8, False, 30],
+            ["src/examples/reference_images/7.jpg", "src/examples/driving_audios/5.wav", "Sad", 1.2, False, 24],
         ],
         inputs=[source_image, driving_audio, emotion_dropdown, cfg_slider, smooth_checkbox, fps_slider],
         outputs=output_video,