What is a booru?

A booru is an image board where every picture is described by tags: short keywords such as long_hair, smile or outdoors. You find pictures by searching tag combinations, and the community keeps the tags accurate. The word is short for Danbooru, the site that gave this style its name. Today it covers that site, the open-source software behind it, and other sites built on similar code or the same idea.

That one idea, tags as the main way to organize and find pictures, explains most of what is odd about boorus: the strange tag names, the search syntax, and why models trained on booru data answer with tag lists instead of sentences. We are not affiliated with Danbooru or any other booru, and we name them only to describe how they work.

Where the word booru comes from

Danbooru's history page says its founder, who goes by Albert, launched the site on May 17, 2005. The name is a real Japanese word: 段ボール (danbōru) means corrugated cardboard, and the site's wiki lists a cardboard box as its namesake. "Booru" is the shortened form communities use for the whole category.

Danbooru's glossary defines it this way: Danbooru is the name of both the website and its software, and sites that use the same software, older or modified versions of it, or software with a similar idea are all called boorus. Wiktionary adds that the ending may echo the Japanese rendering of the English word "board", but it marks that etymology as uncertain.

The style spread because Albert released Danbooru's original source code in March 2007. The history page says this led to many booru-style image boards, some built on the code and some only inspired by it.

How a booru differs from an ordinary photo site

On many photo sites the uploader writes a title and a caption, and people discover pictures through accounts, albums and feeds. On a booru the tags are the interface. They are the main way anyone finds a picture, and on Danbooru any member can change them: its FAQ says anyone with an account can edit a post's tags.

That vocabulary needs constant upkeep. Danbooru's own post counter reads over 12 million as of October 2026, and its history page puts post number 12,000,000 on August 16, 2026.

The main booru software families

Because the code is open, anyone can run a booru. The software decides details that matter later: how tags are written, which ratings exist and how many tags a search may use. All five families below publish their source code on GitHub, and the table comes from their code and settings.

Moebooru is a heavily modified fork of the older Danbooru 1 code, while the Danbooru site itself moved to Danbooru 2 in February 2013. Shimmie2 calls itself an easy-to-install community image gallery, and Szurubooru aims at small and medium communities. Some large sites wrote their own engines and others run old or modified Danbooru code, so two sites that look alike can behave differently.

This guide names software rather than listing individual sites, because boorus differ in what they host and some carry adult material. If you want a safe-for-work view of Danbooru, its Safe mode hides every post not rated general, and the glossary points to donmai.moe for that. Check a site's rules and rating filters before you browse.

How tags work in practice

Danbooru's tag rules are the template many sites follow. The basics, from its tagging guide and help pages:

• Spaces separate tags, so names use underscores: looking_at_viewer.
• Names have limits. They cannot contain commas or asterisks or start with a dash or tilde, and "and" and "or" are reserved because search uses them.
• Parentheses tell look-alikes apart. photo_(medium) describes how a work was made, while photo_(object) marks a photograph inside the frame.

Two mechanisms keep the vocabulary consistent. An alias redirects an alternate name to the main tag, in search and when tagging: yellow_hair is an alias of blonde_hair, and gray_hair of grey_hair. An implication adds a broader tag automatically: denim_shorts adds denim and shorts, and 4koma adds comic. Implications only join tags of the same type, and character tags do not imply their series, because that would be too much to maintain. We checked these examples against Danbooru's live data on October 18, 2026.

Tags can also be retired. A deprecated tag can be removed from posts but never added, which Danbooru does to names that are ambiguous, too broad or subjective. cute is one, with zero posts as of October 18, 2026. For the five tag categories and the counting tags, see our guide to how Danbooru tags work, and for everyday examples the booru tag glossary.

Why tag names differ between sites

Each site is its own community, with its own wiki, aliases and decisions about retiring tags, so the same idea may be spelled differently, and a tag Danbooru has retired may still be normal elsewhere. The software adds the differences in the table above. Tag types vary too: Danbooru has five (general, character, copyright, artist and meta), while Moebooru's default configuration defines four, with no meta type. If you copy tags between sites, expect to fix the separators and the rating and to find some tags aliased or missing.

Searching a booru

Search syntax varies by software, so this section uses Danbooru's, documented in its cheat sheet. A space means AND:

Words with a colon are metatags: they filter on facts about a post, such as its rating, instead of its tags. Use rating:g,s, not two rating terms, because repeating a metatag does not work.

According to Danbooru's user levels page, visitors without an account and free Member accounts can search two tags at a time, Gold accounts six, and Platinum and above have no limit. Many metatags, including rating, score, id and date, do not count. We tested it on October 18, 2026 without an account: blonde_hair blue_eyes rating:g returned results, while adding a third tag returned the error "You cannot search for more than 2 tags at a time." An excluded tag counts too, so pick the two tags that narrow the search most and let the free metatags do the rest. Moebooru's default limit is six.

Ratings: three levels, four levels or your own

Boorus attach a content rating to every post so people can filter what they see. Danbooru has four levels: general, sensitive, questionable and explicit. In short, general is meant to be safe for any audience, sensitive covers suggestive or mildly revealing pictures, and the top two levels hold adult or graphic material, with explicit the strongest. The rating guide spells out each level. Danbooru's history page dates the four-level system to May 22, 2022, when general was added and the old "safe" level was renamed sensitive.

Other software uses three levels or its own scheme, as the table above shows. The lines differ even when names match: what one site calls safe may be general or sensitive on another, so convert ratings deliberately when you move tag data.

If you want only pictures that are completely safe for work, search rating:g, because Danbooru's guide describes sensitive as covering anything potentially not safe for work. And ratings are volunteer judgment calls: the guide says to pick sensitive over general when unsure. WD14 taggers predict the four ratings alongside ordinary tags, and our Danbooru tags guide explains how to keep or drop them.

Good tagging habits

Danbooru's tagging guide and tag checklist boil down to habits that work on any booru and in your own dataset captions:

• Tag what you see, not what you know. If a picture is cropped at the waist, leave out the shoes you know the character wears.
• Leave out opinions. Subjective tags such as cute or good are discouraged because opinions conflict.
• Count the people first. The checklist starts with counters such as 1girl, 2boys and solo, then framing (upper_body, full_body) and viewpoint (from_above, from_side).
• Use one tag per fact. The checklist advises against theme tags when existing ones cover the idea: battle plus blood, not a "bloody fighting" tag.
• Say when you don't know. tagme flags a post that needs more tags, but Danbooru warns against using it as an excuse to skip tagging your own uploads.

Applied to a girl in a red scarf, standing in snow, shown from the waist up and looking at the viewer, the habits give:

1girl solo red_scarf snow outdoors standing upper_body looking_at_viewer

Everything in that line is visible, and Danbooru adds scarf by itself because red_scarf implies it.

Why machine learning models speak booru

Danbooru's millions of volunteer-labeled pictures, all tagged with one documented vocabulary, made it a natural training set. The WD14 taggers behind our tools, trained by SmilingWolf, learned from it. The model card gives the last post ID in the training data as 5,944,504, which Danbooru's API dates to the final minute of December 31, 2022, and says tags with fewer than 600 images were dropped. Some anime image generators lean on the same vocabulary. The Animagine XL 3.1 model card says that model is optimized for Danbooru-style tags instead of natural-language prompts, and its recommended quality prompt includes absurdres, a Danbooru meta tag for very high-resolution images.

A model knows the vocabulary as it stood when training ended, and Danbooru has kept growing and renaming. The newest post number we saw on October 18, 2026 was 12,405,349, so by post number more than half of Danbooru's history came after the WD14 cutoff. 1girl was on 4,225,150 posts when the WD14 v2 tag list was built and on about 8.56 million in Danbooru's live data on that date. Our own snapshot from 2026-10-15 found that of the 9,079 general and character tags the default model can output, 270 are now aliases and 233 are deprecated, about one in eighteen. Models still produce the old names, which is why the Danbooru tagger swaps in current ones. Newer WD taggers exist, so check each model card for its cutoff, and see our WD14 model comparison for the ones we offer.

To turn booru-style tags into training captions, see LoRA dataset captions. Tags versus natural-language captions explains when sentences work better.

What to do next

• Get tags for a picture. The booru tagger writes space-separated tags with underscores and an optional rating tag. The Danbooru tagger also shows post counts and current tag names, and the anime tagger writes a prompt-style list.
• Tag a whole folder. The batch tagger runs WD14 models over many images and saves a .txt caption file for each.
• Learn the vocabulary. Read Danbooru tags explained, with the tag glossary beside you.