Update app.py

app.py

@@ -4,31 +4,35 @@ from diffusers import AutoencoderKLWan, WanPipeline
 from diffusers.utils import export_to_video
 import spaces  # ZeroGPU integration
 
-@spaces.GPU  # Request a GPU via ZeroGPU
-def generate_video(prompt, negative_prompt=""):
+@spaces.GPU  # This decorator will request a GPU during initialization
+def load_pipeline():
     model_id = "Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
-    
-    # Load the VAE and pipeline with proper data types
+    print("Loading model. This may take several minutes...")
     vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
     pipe = WanPipeline.from_pretrained(model_id, vae=vae, torch_dtype=torch.bfloat16)
     pipe.to("cuda")
-    
-    # Generate video frames at 480p resolution (480x832) with desired settings
-    output = pipe(
+    print("Model loaded successfully.")
+    return pipe
+
+# Preload the model during startup
+PIPELINE = load_pipeline()
+
+def generate_video(prompt, negative_prompt=""):
+    # Use the globally preloaded PIPELINE
+    output = PIPELINE(
         prompt=prompt,
         negative_prompt=negative_prompt,
         height=480,        # 480p height
-        width=832,         # Suitable width for 480p
-        num_frames=81,     # Adjust as needed for video length
-        guidance_scale=5.0 # Recommended for T2V-1.3B model
+        width=832,         # Suitable width for 480p videos
+        num_frames=81,     # Adjust number of frames for desired video length
+        guidance_scale=5.0 # Recommended guidance scale for the 1.3B model
     ).frames[0]
     
-    # Export the generated frames to a video file
     video_path = "output.mp4"
     export_to_video(output, video_path, fps=15)
     return video_path
 
-# Create a Gradio interface for video generation
+# Create the Gradio interface
 iface = gr.Interface(
     fn=generate_video,
     inputs=[
@@ -41,4 +45,4 @@ iface = gr.Interface(
 )
 
 if __name__ == "__main__":
-    iface.launch()
+    iface.launch()