A booru is an image board where every picture is described by short, searchable tags, and a booru tagger is a tool that suggests those tags for you. This one looks at a picture and writes out a tag string in the style the big boorus share: lowercase tags separated by spaces, an underscore inside any tag that has more than one word (long_hair, school_uniform), and an optional rating tag at the front. There is nothing to install and no account to make.
People reach for a booru tagger for three jobs. Some want to tag their own artwork or a folder of saved pictures before posting them to an image board. Some want the tag names themselves, to find out what a booru calls a certain outfit or pose before they search. And many are building captions for a model that was trained on booru tags. The sections below cover all three, and say plainly which parts of the output carry over from one booru to another and which you need to check by hand.
Under the hood it is the WD14 tagger by SmilingWolf, which learned from millions of Danbooru images that volunteers had already tagged. Because most other boorus borrowed Danbooru's conventions, those tags are a good starting point on many sites. The model runs in your browser. If you want the version of this page that concentrates on Danbooru itself, see the Danbooru tagger.
How to use it
- Add pictures. Drag them anywhere onto the page, click the box to pick files, drop a whole folder, or paste a copied image with Ctrl+V (Cmd+V on a Mac). Each picture appears at once and fills in when its turn comes.
- Let the model load. On your first picture the browser downloads the tagging model once and keeps a copy, so later pictures only take a few seconds.
- Review the chips. Every tag is a chip with a grey number (how many Danbooru posts use it) and a percentage (how sure the tagger is). Click a chip to leave that tag out. Slide toward "Fewer, surer tags" to shorten the list.
- Copy. The box under the chips holds the finished string. You can edit it by hand, then press Copy tags.
What the output looks like
A typical result for a picture of two friends under cherry trees reads:
rating:g 2girls school_uniform smile outdoors cherry_blossoms standing looking_at_viewer
Four switches under the box change how it is written. Each one is set per picture:
| Switch | Effect | Typical use |
|---|---|---|
| Use Danbooru's current tag names | Replaces tags Danbooru has since renamed and leaves out tags it has retired | On for posting or searching. Off for the model's raw output. |
| Add rating tag | Puts rating:g, rating:s, rating:q or rating:e first | On for Danbooru. Off if the site you post to has its own rating buttons. |
| Include characters | Adds recognized character names, but only when the model is at least 85% sure | Off for a generic description |
| Commas between tags | Writes 1girl, solo, long_hair instead of 1girl solo long_hair | On for caption files and training tools that expect commas. Leave it off when posting to a booru: Konachan's tagging help says commas are stripped from tags. |
Which boorus this suits
"Booru" names a style of site, not one website, and the sites do not all agree on every detail. Danbooru, Gelbooru and the sites built on the Moebooru software (Konachan and yande.re are examples) all separate tags with spaces and write multi-word tags with underscores, so a string from this page pastes into their tag boxes the same way. Leave "Commas between tags" off when you do: Konachan's own tagging help says commas are stripped from tags. The differences that matter are smaller and more annoying:
| Question | What to check |
|---|---|
| Does the site use the same tag names? | Mostly, because the tag vocabulary spread from Danbooru. But each site keeps its own aliases and its own deprecated tags, so a name this page treats as outdated may still be the normal one elsewhere. |
| How is the rating set? | Danbooru has four levels: general, sensitive, questionable and explicit. Sites on the Moebooru software have three: safe, questionable and explicit. Use the rating buttons in the upload form if there are any, and switch the rating tag off here. |
| Are there tags the site expects that the tagger can't write? | Yes, on every site: the artist, the series or copyright, and the source of the picture. See "What the tagger leaves out" below. |
| Do the grey post counts apply? | No. They are Danbooru's own counts, so use them as a rough guide to how established a tag is and nothing more. |
Boorus built around a different subject usually have their own tag vocabulary as well. A site for photographs, or for a particular fandom, will not know most of what the WD14 model says, and the model cannot write tags it was never taught. For those, treat the output as a list of ideas rather than a ready-made string. Our guide to booru sites explains how the different families of sites compare.
What the tagger leaves out
Danbooru sorts tags into five types, and the model only writes two of them. The rest you add yourself:
- General tags describe what is in the picture: hair, eyes, clothes, expression, pose, background. This is nearly everything the tagger produces.
- Character tags name a character. The model knows 2,132 popular ones and only writes a name when it is confident.
- Copyright tags name the series or franchise. A character tag often carries its series in brackets, as in
ganyu_(genshin_impact), but the series tag on its own is yours to add. - Artist tags credit whoever drew the picture. An image cannot tell the model that, so it never guesses.
- Meta tags describe the file or the post, such as
highresortranslated.
This is why a booru tagger is best treated as a strong first draft. It does the tedious part, finding thirty or forty descriptive tags in seconds, and leaves the facts only you know to you.
Renamed and retired tags
The model stopped learning in late 2022, and the tag system it learned from has kept changing. We compared the 9,079 general and character tags the model can output with Danbooru's public tag data on October 15, 2026. 270 are now aliases of a different tag, which means Danbooru swaps the old name for a newer one automatically, and 233 are deprecated, which means Danbooru no longer uses them. With "Use Danbooru's current tag names" on, this page swaps the first group and leaves out the second, and it marks both on the chips so you can see what changed.
| The model says | Danbooru now uses |
|---|---|
baozi | steamed_bun |
china_dress | qipao |
doughnut | donut |
The best-known retired tag is bangs: Danbooru's wiki says the plain tag is no longer used, because it would fit nearly every picture of a person. The model still predicts it on most portraits. If you are posting somewhere other than Danbooru, or captioning for a model trained on older data, the old names may be the right ones, which is why the option is a switch.
A quick review routine
The model is right most of the time and wrong in predictable ways, so a short routine catches most problems. This is the order we use:
- Check the head count. Tags such as
1girl,2girls,1boy,soloandmultiple_girlsare among the most-used on Danbooru (1girlis on about 8.5 million posts). A wrong count is easy to miss and it affects every search that uses it. - Check hair and eyes. Compare them with the picture, because strong lighting and color shifts make hair and eye colors harder to judge for people and models alike.
- Click out the guesses. Clothing and accessory tags are the likeliest to be wrong, especially for small details that the 448-pixel view makes hard to see.
- Read the lowest percentages. Anything near the cut-off deserves a second look, and a quick glance at the picture settles most of them.
- Add what is missing. Artist, series, and any specific detail the model has no word for.
Our guide to confidence thresholds explains how to choose the slider setting, and the tag glossary defines the common tags if one of them is new to you.
Before you post to a booru
- Read that site's rules. Every booru has its own upload rules, and they change. Danbooru's rules do not allow fully AI-generated artwork. Other sites set their own policies, so check before you post.
- Only post what you may post. Use the tagger on your own artwork, on pictures you have permission to share, and on your own dataset. It does not make an image yours to upload.
- Fill in the source. Most boorus expect a link to where the picture came from, and this tool cannot know it.
- This site is not affiliated with any booru. It is an independent tool that uses Danbooru's public tag data.
Booru tags for training and prompts
Many anime image models were trained on pictures captioned with Danbooru tags, which is why a tag list from a picture you like makes a useful starting prompt. For a training set, tick "Commas between tags" and copy each caption into a text file next to its image. For a whole folder, the batch tagger runs the same model on every file and exports one .txt caption per image, and its tag editor can add or remove a tag across the whole set at once.
If your generator wants prompt-style tags, with spaces instead of underscores and escaped brackets, the anime tagger formats them that way. The LoRA captioning guide shows how to clean up and structure a set of captions, and tag pruning covers which tags to remove before training.
Where it makes mistakes
- It was built for drawings. On a photograph it will write sensible generic tags such as
outdoorsandsky, but its vocabulary is made for illustration and anime. - It sees a small picture. Every image is shrunk to 448 by 448 pixels first, so small details disappear. Crop to the part you care about.
- Its knowledge stops in 2022. The training data runs to the end of 2022, so newer characters and outfits get older, more generic tags.
- Percentages are not proof. A tag at 90% is nearly always right. One at 40% often is not.
Frequently asked questions
What is a booru tagger?
It is a tool that looks at a picture and suggests the tags a booru would give it. A booru is an image board where volunteers label every post with short tags, so a model trained on those posts can describe a new picture in the same style.
Which boorus can I use these tags on?
The tag style (lowercase, spaces between tags, underscores inside them) is shared by Danbooru, Gelbooru and sites built on the Moebooru software such as Konachan and yande.re, so the string pastes into their tag boxes. The tag names come from Danbooru, though, so each site may treat some of them differently. Check the site's own help for its rules and its rating system.
Is this the tagger Danbooru or Gelbooru uses?
No. This site is not affiliated with any booru. It uses the WD14 tagger model, which was trained on Danbooru data, and Danbooru's public tag information for the post counts and for the renamed and retired tag checks.
What is the difference between this page and the Danbooru tagger?
They run the same tagger and write the same output. This page is written for people tagging for boorus in general, and the Danbooru tagger page concentrates on Danbooru's own format and rules. Use whichever matches what you are doing.
Can I use it for Stable Diffusion or LoRA captions?
Yes. Tick "Commas between tags" for a caption-file style list. For a whole folder use the batch tagger, which saves one caption file per image. The LoRA captioning guide covers cleaning the captions afterwards.
Why does it add a rating tag, and can I turn it off?
Danbooru gives every post a rating, and its short forms are rating:g, rating:s, rating:q and rating:e (general, sensitive, questionable and explicit). The tagger guesses one from the picture. Untick "Add rating tag" if the site you post to uses a different rating system or has its own rating buttons. Always check the guess yourself, because ratings are a judgment call.
Does it tag the artist or the series?
No. It writes general tags and, when it is sure, a character name. It never writes artist, series or source tags, since a picture cannot tell the model who drew it. Add those by hand before you post anywhere that expects them.
Can I tag a whole folder?
You can drop up to 100 pictures on this page at once. For bigger sets, or to save a caption file for every image, use the batch tagger, which runs the same model.