diff --git a/unsloth/models/vision.py b/unsloth/models/vision.py
index e741c1c628..d083144651 100644
--- a/unsloth/models/vision.py
+++ b/unsloth/models/vision.py
@@ -53,9 +53,10 @@ def _wrap_fast_inference(generate, device_type, dtype, model):
 
         kwargs["pad_token_id"] = kwargs.pop("pad_token_id", model_eos_token_id)
 
-        if "pixel_values" in kwargs:
+        try:
             kwargs["pixel_values"] = kwargs["pixel_values"].to(model.dtype)
-        pass
+        except:
+            pass
 
         # Autocasted
         with torch.autocast(device_type = device_type, dtype = dtype):