diff --git a/rag/llm/cv_model.py b/rag/llm/cv_model.py
index fbccd5dff..5303dfca0 100644
--- a/rag/llm/cv_model.py
+++ b/rag/llm/cv_model.py
@@ -629,7 +629,7 @@ class GeminiCV(Base):
         from PIL.Image import open
 
         b64 = self.image2base64(image)
-        vision_prompt = self.vision_llm_prompt(b64, prompt) if prompt else self.vision_llm_prompt(b64)
+        vision_prompt = prompt if prompt else vision_llm_describe_prompt()
         img = open(BytesIO(base64.b64decode(b64)))
         input = [vision_prompt, img]
         res = self.model.generate_content(