diff --git a/vlmeval/vlm/llava/llava.py b/vlmeval/vlm/llava/llava.py
index 2099d02f..3f350b13 100644
--- a/vlmeval/vlm/llava/llava.py
+++ b/vlmeval/vlm/llava/llava.py
@@ -815,12 +815,12 @@ def generate_inner_image(self, message, dataset=None):
             {
                 "role": "user",
                 "content": [
-                    {"type": "text", "text": content.split("\n", 1)[-1]},
-                    {"type": "image"},
+                    {"type": "text", "text": content},
                 ],
             }
         ]
         prompt = self.processor.apply_chat_template(conversation, add_generation_prompt=True)
+        print(prompt)
         inputs = self.processor(images=images, text=prompt, return_tensors="pt").to('cuda', torch.float16)
 
         output = self.model.generate(**inputs, max_new_tokens=512)