None of the models use images, they all operate and a json format the describes the input squares.