Visualizing Source Code as Comics Using Generative AI

Abstract

The architecture and inner structure of software is often only implicitly available in the form of its source code and thus not tangible and intuitively easy to understand for non-programmers and laymen. Our goal is to create visualizations as automatically as possible, with which such people can nevertheless understand the software or parts of the software and get a feel for the structure of the software and how its methods work. Especially for newcomers to software projects, for management or even for students and pupils, it can be helpful to get a non-technical insight into the software.

We use the concept of visualizing information as comics to present aspects of the software as strikingly as possible, as comics are an effective way to present complex systems and interrelationships for certain target groups. For this purpose, we present a method to generate comics from source code. Our semi-automated process is based on generating a prompt for an LLM from source code, which in turn generates a prompt for a comic image generation using the text-to-image model Stable Diffusion. We show that generative AI methods can be used to rapidly generate human-compatible artistic representations from source code. However, further research is needed to validate the understandability of the results.