MCP server ready

hf_gradio_app.py

@@ -102,6 +102,23 @@ def process_audio(file_path, temp_dir):
 @spaces.GPU(duration=240)
 @torch.inference_mode()
 def generate(input_video, input_audio, seed, progress=gr.Progress(track_tqdm=True)):
+    """
+    Generates a talking-head video synchronized with the input audio using the MEMO pipeline.
+
+    This function combines an input face image and an audio clip to create a temporally coherent 
+    and emotionally expressive talking video. It leverages a memory-guided diffusion model 
+    conditioned on audio features, emotional cues, and visual context.
+
+    Args:
+        input_video (str): Path to the input image file (used as the reference face).
+        input_audio (str): Path to the input audio file (speech or dialogue).
+        seed (int): Random seed for deterministic results. Use 0 for a randomly generated seed.
+        progress (gr.Progress, optional): Gradio progress tracker (automatically passed by Gradio).
+
+    Returns:
+        str: File path to the generated output video (MP4 format).
+    """
+    
     pipeline.reference_net.enable_xformers_memory_efficient_attention()
     pipeline.diffusion_net.enable_xformers_memory_efficient_attention()
 
@@ -246,4 +263,4 @@ with gr.Blocks(analytics_enabled=False) as demo:
         outputs=[video_output],
     )
 
-demo.queue().launch(share=False, show_api=False, show_error=True)
+demo.queue().launch(share=False, show_api=True, show_error=True, mcp_server=True)