Autonomous Generation of 2-panel Comics Based on Human-made Comics

Abstract

This paper reports an AI has made 2-panel comics using rules of meta-information obtained from human-made 4-panel comics. The 4-panel comic consists of 4 panels, and each panel consists of characters, backgrounds, balloon texts, onomatopoeia, and cartoon marks. At the interesting point, we can restore a comic strip which was broken up into panels separately. This means that there is something about connecting in the panels. Then, this paper describes that the experiment to find such a connection rule and make a new story by combining selected panels. The analysis led the rules for dialogue and characters. As a result, the AI made stories up to 2 panels without collapse. This experiment indicated the effectiveness of abstracting different comics and creating new comics from common elements. In the future, this AI will be expected to try out not only 4-panel comics but also story comics using the similar method.