<fix> seperate model init and inference.

app.py

@@ -46,9 +46,9 @@ The corresponding condition images will be automatically extracted.
 """
 
 
+def init_pipeline():
+    global pipe
 
-@spaces.GPU
-def process_image_and_text(condition_image, target_prompt, condition_image_prompt, task):
     # init models
     transformer = HunyuanVideoTransformer3DModel.from_pretrained('hunyuanvideo-community/HunyuanVideo-I2V', 
                                                                  subfolder="transformer", 
@@ -162,6 +162,10 @@ def process_image_and_text(condition_image, target_prompt, condition_image_promp
         image_processor=image_processor,
     )
 
+
+@spaces.GPU
+def process_image_and_text(condition_image, target_prompt, condition_image_prompt, task):
+
     # start generation
     c_txt = None if condition_image_prompt == "" else condition_image_prompt
     c_img = condition_image.resize((512, 512))
@@ -235,6 +239,9 @@ def process_image_and_text(condition_image, target_prompt, condition_image_promp
             c_img = c_img.resize((args.img_size, args.img_size))
             c_img.save(os.path.join(save_dir, f"low_to_high.png"))
 
+    if pipe is None:
+        init_pipeline()
+
     gen_img = pipe(
         image=c_img,
         prompt=[t_txt.strip()],