up

app.py

@@ -14,15 +14,13 @@ logger = logging.getLogger(__name__)
 # Constants
 DEFAULT_CONFIG_PATH = "configs/inference.yaml"
 DEFAULT_INPUT_FILE = "examples/infer_samples.txt"
-OUTPUT_DIR = Path("demo_out/gradio_outputs")
-OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
 
 def generate_avatar_video(
     reference_image,
     audio_file, 
     text_prompt,
     seed=42,
-    num_steps=50,
+    num_steps=15,
     guidance_scale=4.5,
     audio_scale=None,
     overlap_frames=13,
@@ -152,15 +150,18 @@ def generate_avatar_video(
             # Get the most recent video
             latest_video = max(generated_videos, key=lambda p: p.stat().st_mtime)
             
-            # Copy to output directory with unique name
-            output_filename = f"avatar_video_{os.getpid()}_{torch.randint(1000, 9999, (1,)).item()}.mp4"
-            output_path = OUTPUT_DIR / output_filename
+            # Create a temporary file for the output video
+            # This file will persist beyond the context manager since we're using delete=False
+            with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp_output:
+                output_path = tmp_output.name
+                
+            # Copy the generated video to the temporary file
             shutil.copy(latest_video, output_path)
             
             progress(1.0, desc="Generation complete")
-            logger.info(f"Video saved to: {output_path}")
+            logger.info(f"Video saved to temporary path: {output_path}")
             
-            return str(output_path)
+            return output_path
             
     except Exception as e:
         logger.error(f"Error generating video: {str(e)}")
@@ -201,25 +202,25 @@ with gr.Blocks(title="OmniAvatar - Lipsynced Avatar Video Generation") as app:
                 with gr.Row():
                     seed = gr.Slider(
                         label="Seed",
-                        minimum=0,
+                        minimum=-1,
                         maximum=2147483647,
                         step=1,
-                        value=42
+                        value=-1
                     )
                     
                     resolution = gr.Radio(
                         label="Resolution",
                         choices=["480p", "720p"],
-                        value="720p"
+                        value="480p"
                     )
                 
                 with gr.Row():
                     num_steps = gr.Slider(
                         label="Inference Steps",
                         minimum=10,
-                        maximum=100,
-                        step=5,
-                        value=50
+                        maximum=50,
+                        step=1,
+                        value=15
                     )
                     
                     guidance_scale = gr.Slider(