Danbooru is an English language imageboard and image hosting website focused primarily on anime style illustrations. It was launched in 2005 by a programmer known as "Albert" and is frequently described as one of the earliest and most influential "booru" style sites, using collaborative tags to organize images.[1][2]
The site hosts over 10 million user submitted images (as of December 2025), which are labeled with user generated tags and content ratings. Danbooru has served as the basis for a large ecosystem of derivative software, related imageboards and machine learning datasets.[3][4]
History
Danbooru was created in 2005 as an imageboard for sharing and cataloging anime and manga style artwork with more structured metadata tagging system than traditional forums or boards.[1][5] According to a later retrospective based on the site's Facebook page and early web archives, Danbooru launched on May 25, 2005 and had tens of thousands of images within its first year of operation.[1] The website's name derives from the Japanese word danbōru (段ボール; lit. "corrugated cardboard").[6]
In March 2007 the site was temporarily taken offline by its administrator, who also released a copy of the database and the site's source code. This closure encouraged other developers to launch their own boards using the Danbooru codebase, leading to a rapid proliferation of so-called "booru"-style imageboards dedicated to particular fandoms or content ratings.[7] Danbooru itself returned later in 2007 and has remained online since.
Over time the site's code base was extracted into a more general purpose open source project and moved to public version control hosting, where it has been maintained and extended as a Ruby on Rails application.[8] Danbooru's operators have also periodically released database exports and have allowed external mirrors for research and archival purposes.[9]
In the 2010s and 2020s, Danbooru became widely referenced in online culture and in technical literature as the archetypal anime imageboard, with spin-offs including sites such as Gelbooru, Derpibooru, E621 and others built on similar tagging conventions.[1][2]
Software
The Danbooru website runs on a custom imageboard engine written in Ruby on Rails and released as free and open-source software under a BSD-style license.[8] The software provides features typical of "booru" platforms, including:
Uploading and hosting of raster images and animations.
A collaborative tagging interface where users can add, edit and imply tags.
Tag type categorization (such as copyright, character, artist, general content and meta tags).
Rating fields distinguishing safe, borderline and explicit content.
A wiki system for documenting tags, series, characters and site specific conventions.
User accounts, favorites, comments and voting.
A forum for users to collaborate.
"Pools", which allow ordered sequences or thematic collections of images.
Because the codebase is openly licensed, it has been forked or adapted by numerous other imageboard projects, and is often cited in documentation for imageboard search engines and third-party taggers.[1][10]
Tagging and content
Danbooru is organized around a detailed, user-maintained tag system. Tags are grouped into broad categories such as artists, characters, copyrights (franchises or works), general descriptive tags and meta information such as medium or resolution. [9] This structure forms a folksonomy that allows users to search for highly specific combinations of attributes, such as particular characters, clothing, poses or visual motifs.
Images are also given an explicitness rating, typically described as safe, sensitive, questionable or explicit, which reflects the presence of nudity, sexual themes or other potentially sensitive material such as violence or drug use.[1][9] In contrast to some other platforms, Danbooru does not attempt to prohibit such material outright. Instead, the website relies on tagging and ratings so that users and downstream tools can filter content according to their preferences. Options include using the Safebooru variant of the website, or using the blacklist function under their account settings.
The site enforces submission guidelines intended to maintain a minimum level of technical quality. Low resolution or heavily compressed images, near-duplicates and certain kinds of prohibited content may be rejected or removed by moderators, and the tagging community is encouraged to correct metadata and add translations where appropriate.[9]
Use in datasets and machine learning
Due to its scale and detailed annotations, Danbooru has been widely used as a source for datasets in computer vision and generative modeling research. The Danbooru20xx series of datasets (including Danbooru2017 and Danbooru2021) are static snapshots of the site's content and metadata prepared for machine learning, comprising millions of images with hundreds of millions of tags.[9][11]
The Danbooru2021 release, for example, includes more than 4.9 million images with roughly 162 million tag annotations, covering uploads from 2005 to the end of 2021.[9] These datasets have been used to train models for multi-label image classification, automatic tagging, anime-style face detection and segmentation, and image generation, among other tasks.[9][12][13]
Later work has extended the concept to larger derivatives such as "Danbooru2023" and related webp based releases, which continue to be cited in research on diffusion models and text to image systems that specialize in anime style artwork.[14]
NovelAI, an AI image generation platform, uses a custom version of the Stable Diffusion text-to-image diffusion model called NovelAI Diffusion, which is reported to be trained on a Danbooru based dataset.[15]
Reception and influence
Danbooru is widely described in secondary sources as a foundational or "canonical" anime imageboard, notable for its emphasis on high-quality submissions and detailed tagging.[4][13] The site's tagging conventions, rating scheme and API have been emulated by numerous subsequent imageboards and related tools, and have influenced how anime and fan-art communities organize visual content online.[1][2]
In addition to its role in online fandom, Danbooru has been used as a case study in discussions of collaborative content curation, crowdsourced annotation and the ethics of large-scale web scraping for machine learning. Academic and technical articles frequently cite Danbooru both as an example of a user-generated tagging system and as a source of training data for models that operate on illustration-style images.[3][12][14][13]
12Xiang, Yuejia; Lv, Chuanhao; Liu, Qingdazhu; Yang, Xiaocui; Liu, Bo; Ju, Meizhi (November 16, 2022). "A Creative Industry Image Generation Dataset Based on Captions and Sketches". arXiv:2211.09035 [cs.CV]. Danbooru2021 is a large-scale anime image database with 4.9m+ images annotated with 162m+ tags.
↑"插畫AI又陷爭議!Novel AI官方自爆以非法搬運網站當訓練來源 生成圖片竟自帶簽名?". United Daily News (in Chinese (Taiwan)). October 4, 2022. Novel AI 近期在官方推特留言區回覆網友留言時, 提及自家人工智慧有透過名叫「Danbooru」的 ACG 圖片網站進行訓練... 許多圖片多來自 pixiv 跟推特的繪師二創、原創作畫... 隨著爭議引爆, 消息也傳到日本, 引起創作者的關切.[NovelAI recently responded to an online commentator's message on its official Twitter, stating that the AI was trained through the ACG image website "Danbooru"... Many pictures are reposts of fanarts and original artworks from Pixiv and Twitter... As the controversy unfurled, the news also reached Japan, causing concern among content creators.]
This page is based on this Wikipedia article Text is available under the CC BY-SA 4.0 license; additional terms may apply. Images, videos and audio are available under their respective licenses.
