A Softer World, a webcomic characterized by its subtly dark and surreal humor, presents a unique challenge for text generation algorithms, as its style is difficult to replicate. This project explores the implementation of a CNN/RNN architecture and a pipeline for it to generate new A Softer World comics. Our algorithm, which inputs any background image and outputs A Softer World-eqsue text relevant to the image, performs significantly better than baseline algorithms such as Markov chains. Furthermore, we have contributed a full dataset of A Softer World comic transcriptions, with text per panel for every 3-panel comic.