If you are able to reveal something of the architecture, is this just classifying the image based off other similar images from r/roastme, and then outputting a random insult that was deemed acceptable for that classification of image? Like there's no NLP happening here, correct?