Update app.py

app.py

@@ -113,31 +113,29 @@ initialize_infer_state(ArgsNamespace())
 # Global Pipeline Variable
 pipe = None
 
-def pre_load_model():
-    """Loads the model into memory/GPU before UI launch."""
-    global pipe
-    
-    # Double check path exists
-    if not os.path.isdir(MODEL_DIR):
-        print(f"❌ Error: Model directory not found at {MODEL_DIR}")
-        sys.exit(1)
+# Double check path exists
+if not os.path.isdir(MODEL_DIR):
+    print(f"❌ Error: Model directory not found at {MODEL_DIR}")
+    sys.exit(1)
 
-    print(f"⏳ Initializing Pipeline ({TRANSFORMER_VERSION}) from {MODEL_DIR}...")
-    
-    try:
-        pipe = HunyuanVideo_1_5_Pipeline.create_pipeline(
-            pretrained_model_name_or_path=MODEL_DIR,
-            transformer_version=TRANSFORMER_VERSION,
-            enable_offloading=ENABLE_OFFLOADING,
-            enable_group_offloading=ENABLE_OFFLOADING,
-            transformer_dtype=DTYPE,
-        )
-        print("✅ Model loaded successfully!")
-    except Exception as e:
-        print(f"❌ Failed to load model: {e}")
-        import traceback
-        traceback.print_exc()
-        sys.exit(1)
+print(f"⏳ Initializing Pipeline ({TRANSFORMER_VERSION}) from {MODEL_DIR}...")
+
+try:
+    pipe = HunyuanVideo_1_5_Pipeline.create_pipeline(
+        pretrained_model_name_or_path=MODEL_DIR,
+        transformer_version=TRANSFORMER_VERSION,
+        enable_offloading=ENABLE_OFFLOADING,
+        enable_group_offloading=ENABLE_OFFLOADING,
+        transformer_dtype=DTYPE,
+    )
+    print("✅ Model loaded successfully!")
+except Exception as e:
+    print(f"❌ Failed to load model: {e}")
+    import traceback
+    traceback.print_exc()
+    sys.exit(1)
+
+pipe.to("cuda")
 
 def save_video_tensor(video_tensor, path, fps=24):
     if isinstance(video_tensor, list): video_tensor = video_tensor[0]
@@ -163,6 +161,8 @@ def generate(input_image, prompt, length, steps, shift, seed, guidance):
     print(f"Generating: {prompt} | Seed: {seed}")
 
     try:
+        pipe.execution_device = torch.device("cuda")
+
         output = pipe(
             prompt=prompt,
             height=480, width=854, aspect_ratio="16:9",
@@ -213,9 +213,6 @@ def create_ui():
     return demo
 
 if __name__ == "__main__":
-    # 1. Execute the pre-load BEFORE the UI launches
-    pre_load_model()
-    
     # 2. Launch UI
     ui = create_ui()
     ui.queue().launch(server_name="0.0.0.0", share=True)