diff --git a/vllm/v1/engine/mm_input_mapper.py b/vllm/v1/engine/mm_input_mapper.py
index 218724bff6bba..8bfc739b3dbbc 100644
--- a/vllm/v1/engine/mm_input_mapper.py
+++ b/vllm/v1/engine/mm_input_mapper.py
@@ -180,6 +180,10 @@ def hash_prompt_mm_data(self, prompt: PromptType) -> Optional[List[str]]:
             return None
 
         mm_data = prompt["multi_modal_data"]
+        if not mm_data:
+            # mm_data can be None or an empty dict.
+            return None
+
         image_inputs = mm_data["image"]
 
         return self.hash_images(image_inputs)