Anybooru

Synthetic Anime Image Dataset

Synthetic anime image dataset generated using Andite's Anything-v4.5 checkpoint with Danbooru2021 tags collected in https://gwern.net/danbooru2021. See https://github.com/1lint/anybooru for details and code to generate your own variant of the dataset. I have also uploaded the extracted Danbooru2021 tags to https://huggingface.co/datasets/lint/danbooru_tags, this dataset was generated with a small subset of the tags in 2021_0_pruned.parquet.

Each string of tags was used to generate 4 different images with different seeds. This serves a similar purpose as random resize crop and image flip transformations to train the model to focus on general concepts encoded in the tags, rather than memorizing specific images.

Quick Start

from datasets import load_dataset

dataset = load_dataset('lint/anybooru')

sample = dataset['train'][0]

image = sample['image']
tags = image.info['tags']

print(tags)

Samples

Each row of samples share the same generation prompt (string of tags).

Citations

@misc{danbooru2021, author = {Anonymous and Danbooru community and Gwern Branwen}, title = {Danbooru2021: A Large-Scale Crowdsourced and Tagged Anime Illustration Dataset}, howpublished = {\url{https://gwern.net/danbooru2021}}, url = {https://gwern.net/danbooru2021}, type = {dataset}, year = {2022}, month = {January}, timestamp = {2022-01-21}, note = {Accessed: 03/01/2023} }