Create app.py (#1)

app.py

@@ -0,0 +1,47 @@
+import os
+import shutil
+from huggingface_hub import snapshot_download
+import gradio as gr
+from scripts.inference import inference_process
+import argparse
+
+# Download the repository contents into a directory
+hallo_dir = snapshot_download(repo_id="fudan-generative-ai/hallo")
+
+# Define the new directory path for the pretrained models
+new_dir = 'pretrained_models'
+
+# Ensure the new directory exists
+os.makedirs(new_dir, exist_ok=True)
+
+# Move all contents from the downloaded directory to the new directory
+for filename in os.listdir(hallo_dir):
+    shutil.move(os.path.join(hallo_dir, filename), os.path.join(new_dir, filename))
+
+def run_inference(source_image, driving_audio, gr.Progress(track_tqdm=True)):
+    # Construct the argparse.Namespace object with all necessary attributes
+    args = argparse.Namespace(
+        config='configs/inference/default.yaml',  # Adjust this path as necessary
+        source_image=source_image.name,
+        driving_audio=driving_audio.name,
+        output='output.mp4',  # You might want to manage output paths dynamically
+        pose_weight=1.0,
+        face_weight=1.0,
+        lip_weight=1.0,
+        face_expand_ratio=1.2,
+        checkpoint=None  # Adjust or set this according to your checkpointing strategy
+    )
+    
+    # Call the imported function
+    inference_process(args)
+    
+    # Return output or path to output
+    return 'output.mp4'  # Modify based on your output handling
+
+iface = gr.Interface(
+    fn=run_inference,
+    inputs=[gr.inputs.Image(type="file"), gr.inputs.Audio(type="file")],
+    outputs="text"
+)
+
+iface.launch()