This article is automatically translated.
Sasaki and Hyoda from DMV attended the “Comic Computing Symposium 2025” held at Ritsumeikan University’s Osaka Ibaraki Campus on November 8, 2025.
Comic Computing is a research field dedicated to the engineering application of manga. Manga is a medium where images and text are complexly interrelated, and research is being conducted across various fields such as image processing, databases, artificial intelligence, interfaces, and cognitive science. The Special Interest Group on Comic Computing (SIG-CC) is an interdisciplinary research group that covers cognitive science and social science related to comic computing, including media associated with manga such as anime. It is open to a wide range of participants, not only researchers but also artists who actually create comics and practitioners from companies.
In this symposium, a keynote speech was given on the digital archiving of media art works, including games. Additionally, lightning talks and discussions were held covering manga from multiple perspectives.
As part of our initiatives regarding manga, DMV introduced “MangaLens,” a manga understanding data infrastructure using Multimodal LLMs, and “OZManga,” a manga advertising video creation software.
MangaLens is a system that extracts information such as story development, character traits, and the worldview of the work from manga page images using Large Language Models (LLMs). By using the pre-extracted content of the manga as context, we anticipate the generation of advertisements based on the content itself and the construction of highly accurate manga recommendation systems.
OZManga focuses on the workflow of creating manga video ads, aiming to be a tool that allows for the easy creation of advertising videos. By utilizing the recursive structure of manga panel layouts, it realizes an intuitive panel-splitting UI.
Please see the slides below for details on the presentation (in Japanese).
Through this symposium, I felt firsthand that the recent evolution of LLMs and foundation models is indeed having a major impact on comic computing. While LLMs and foundation models demonstrate high capabilities in the recognition and generation of images and text, I also felt that there are still many challenges remaining for their application to manga.
Manga, in particular, is a very difficult medium to handle as “data.” It consists of a vast amount of image data, has complex layouts of drawings and text, and above all, the stories are long and contain ambiguity in interpretation.
In thinking about systems that handle manga content, I reaffirmed that the central issue is “how we should design the extraction of features from the content.” Also, from the perspective of digital archiving—which was also the theme of the keynote speech—I was impressed by the emphasis on the importance of not just digitizing and preserving works, but making them into a “usable format.” If a data infrastructure that allows for content-based searching and analysis of manga is established, the possibilities for academic use will expand greatly.
Furthermore, manga is a composite medium involving not only authors but also publishers and many rights holders. Therefore, even when advancing technical research, I felt it is extremely important for industry and academia to collaborate and mutually understand rights management and the realities of the production environment. I am looking forward to the next event.
Kazuma Sasaki
