diff --git a/vllm/model_executor/models/gemma3_mm.py b/vllm/model_executor/models/gemma3_mm.py
index d14f5fa3d608b..d756f54c49b0e 100644
--- a/vllm/model_executor/models/gemma3_mm.py
+++ b/vllm/model_executor/models/gemma3_mm.py
@@ -627,7 +627,7 @@ class Gemma3ForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP,
 
             inputs_embeds = self.get_input_embeddings(input_ids,
                                                       vision_embeddings)
-            if vision_embeddings is not None:
+            if (vision_embeddings is not None) and len(vision_embeddings) != 0:
                 kwargs = self.prepare_attn_masks(
                     input_ids,
                     positions,