The MangaZero dataset is a large-scale, multi-role, multi-state comic image dataset designed for comic generation tasks. It was released by Peking University, Shanghai Artificial Intelligence Laboratory, Nanyang Technological University and ByteDance in 2024. The relevant paper results are "DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga GenerationSpecifically, it contains 43,264 pages of comics and 427,147 annotated panels, which supports the visualization of various character interactions and actions in consecutive frames and is suitable for multi-character and multi-state comic generation tasks.

Citation

article{wu2024diffsensei,
title={DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga Generation},
author={Jianzong Wu, Chao Tang, Jingbo Wang, Yanhong Zeng, Xiangtai Li, and Yunhai Tong},
journal={arXiv preprint arXiv:2412.07589},
year={2024},
}