Is it possible to use this model to caption one image in code, for example, in google colab, without running script?