Paper published on a website (Scientific congresses and symposiums)
A Deep Learning Pipeline for the Synthesis of Graphic Novels
Manouach, Ilan; Melistas, Thomas; Siglidis, Yannis et al.
2021International Conference of Computational Creativity (ICCC 21)
Peer reviewed


Full Text
Author postprint (3.67 MB)

All documents in ORBi are protected by a user license.

Send to


Keywords :
computational creativity; manga; comics; machine learning
Abstract :
[en] In this paper, we present what is to the best of our knowledge, the first deep learning pipeline to produce a synthetic graphic novel. Our method can synthesize from scratch engaging sequences of graphic novel pages, focusing on the Manga genre. To achieve this, we extract images and text from around 670 thousand Manga pages, which we use separately in order to train state-of-the-art generative architectures, such as GPT-2 for text generation and StyleGAN2 for image synthesis. Using these as sources of synthetic content, we develop a set of algorithmic aesthetic rules in order to bring together complete and continuous Manga pages.
Disciplines :
Computer science
Author, co-author :
Manouach, Ilan  ;  Université de Liège - ULiège > Traverses
Melistas, Thomas
Siglidis, Yannis
Kalogiannis, Fivos
Language :
Title :
A Deep Learning Pipeline for the Synthesis of Graphic Novels
Publication date :
13 August 2021
Event name :
International Conference of Computational Creativity (ICCC 21)
Event organizer :
Association for Computational Creativity
Event date :
September 14-18, 2021
By request :
Audience :
Peer reviewed :
Peer reviewed
Available on ORBi :
since 23 May 2024


Number of views
4 (0 by ULiège)
Number of downloads
1 (0 by ULiège)


Similar publications

Contact ORBi