Skip to content

Dataset · jianzongwu

Manga character annotations from the DiffSensei paper

jianzongwu/MangaZero

Text annotations of manga characters with face classification for comic generation.

What's inside

The MangaZero dataset (from 10 to 100 thousand records) is from the DiffSensei research paper dedicated to custom manga generation using multimodal LLMs and diffusion models. The annotations use a type key dividing characters by appearance type: clear faces (0), blurred faces (1), and exaggerated artistic images (2).

What you can do with it

  • Training manga generation models with character appearance preservation
  • Filtering training data by face drawing quality and style
  • Reproducing experiments from the DiffSensei paper
Tasks
Text → image
Size
10K–100K rows
Downloads this month
168
Likes
37

How to load it

from datasets import load_dataset

ds = load_dataset("jianzongwu/MangaZero")