Yes, it's a "zero-shot" deep learning model, it's using both image and the text from the page to make predictions.