better commenting on model loading on zeroGPU or not if cuda not available

app.py

@@ -40,6 +40,16 @@ try:
         raise FileNotFoundError(f"Base model not found at: {BASE_MODEL_PATH}")
     if not os.path.exists(LORA_MODEL_PATH):
         raise FileNotFoundError(f"LoRA model not found at: {LORA_MODEL_PATH}")
+
+    print(f"Is CUDA available: {torch.cuda.is_available()}")
+    if torch.cuda.is_available():
+        face_processor = FaceProcessor(antelopv2_path=ANTELOPEV2_PATH)
+        pipe = load_wan_pipe(base_path=BASE_MODEL_PATH, torch_dtype=torch.bfloat16)
+        set_stand_in(pipe, model_path=LORA_MODEL_PATH)
+        print("Model loaded successfully!")
+    else:
+        print("Will load models on ZeroGPU on inference if available")
+
  
 except Exception as e:
     print(f"Model loading failed: {e}")
@@ -56,7 +66,7 @@ except Exception as e:
     demo.launch()
     exit()
 
-@spaces.GPU(duration=280)
+@spaces.GPU(duration=300)
 def generate_video(
     pil_image: Image.Image,
     prompt: str,
@@ -70,17 +80,22 @@ def generate_video(
     if pil_image is None:
         raise gr.Error("Please upload a face image first!")
 
-    face_processor = FaceProcessor(antelopv2_path=ANTELOPEV2_PATH)
-    pipe = load_wan_pipe(base_path=BASE_MODEL_PATH, torch_dtype=torch.bfloat16)
-    set_stand_in(pipe, model_path=LORA_MODEL_PATH)
-    print("Model loaded successfully!")
+    if is_shared_ui:
+        print("Loading models...")
+        face_processor = FaceProcessor(antelopv2_path=ANTELOPEV2_PATH)
+        pipe = load_wan_pipe(base_path=BASE_MODEL_PATH, torch_dtype=torch.bfloat16)
+        set_stand_in(pipe, model_path=LORA_MODEL_PATH)
+        print("Model loaded successfully!")
 
     print("Processing face...")
     ip_image = face_processor.process(pil_image)
     print("Face processing completed.")
 
     if is_shared_ui: 
-       num_steps = 10
+        num_steps = 10
+        quality = 5
+        fps = 12
+        
     
     print("Generating video...")
     start_time = time.time()
@@ -171,11 +186,11 @@ with gr.Blocks() as demo:
                     info="More steps may improve details but will take longer to generate.",
                 )
                 output_fps = gr.Slider(
-                    label="Video FPS", minimum=10, maximum=30, step=1, value=25,
+                    label="Video FPS", minimum=10, maximum=30, step=1, value=12,
                     interactive = False if is_shared_ui else True,
                 )
                 output_quality = gr.Slider(
-                    label="Video Quality", minimum=1, maximum=10, step=1, value=9,
+                    label="Video Quality", minimum=1, maximum=10, step=1, value=6,
                     interactive = False if is_shared_ui else True
                 )