Published October 27, 2026by Mexes

Rule34 Tag Taxonomy - Open Source Dataset

213 views3 reactions0 comments on CivitAI7 collected

datasetannouncementtagdatataxonomy

DATASET: https://huggingface.co/datasets/Mexes/rule34-tag-taxonomy

DANBOORU VERSION: https://huggingface.co/datasets/Mexes/danbooru-tag-taxonomy

BOORU PROMPT GALLERY: https://booru-prompt-gallery.vercel.app/

TL;DR: The Rule34 companion of the Danbooru Tag Taxonomy, released as a free, open source dataset. It has 13,444 Rule34 tags that aren't in the Danbooru taxonomy, each sorted into a category and one of the same 33 slots (hair, legwear, expression, setting...). An AI sorted them and no person checked them, so every tag says whether the AI was sure. CC BY 4.0, so use it for whatever you want.

Hello! A few weeks ago I released the Danbooru Tag Taxonomy, the categories behind Booru Prompt Gallery. Rule34 has a lot of vocabulary Danbooru doesn't. It writes brunette where Danbooru writes brown_hair, and it has a whole furry vocabulary (anthro, canine, scalie...). Booru Prompt Gallery needs to understand those tags to write sentences from Rule34 posts, so I sorted them the same way.

What is this?

Every tag gets a category and a slot from the same taxonomy as the Danbooru one. stockings is clothing:legwear, cherry_blossoms is scenery:atmosphere and domestic_cat is creature:animal. As in the Danbooru one, tags in the same slot can usually replace each other.

Each tag also has a status.

approved: the AI was at least 70% sure. 8,191 tags.
needs_review: it wasn't, so the tag is parked in other:unclassified. 5,253 tags. For 3,794 of them the AI's low-confidence guess is still there in proposed_slot, in case you want to check them yourself.

Rule34 tags are harder to sort than Danbooru's, because many are very specific and many only make sense if you know the site.

Why is it a separate dataset?

It's less curated than the Danbooru one, and I don't want to mix the two.

No human review. The Danbooru taxonomy was cross-checked and fixed by hand where people reported mistakes. The Rule34 slots are the AI's unchecked answers.
Tags that mean something else. Some Rule34 tags have a meaning of their own on the site. On Rule34, gloves_(marking) are fur markings shaped like gloves on an animal character, but the AI put them in clothing:accents with 93% confidence, and there will be more cases like that.
Messier vocabulary. Rule34 keeps near-duplicates and misspellings as separate tags (canine, canid and canis, or racoon_girl next to raccoon_girl).

If you only want the reliable part, keep the approved tags, and still expect some mistakes.

Why did I make it public?

I needed these slots for my app anyway, and anyone who works with Rule34 tags runs into the same vocabulary. I think they can be useful for a lot more than Booru Prompt Gallery, and I'm sure people will come up with better uses than mine.

How was it made?

Tags: Rule34 general tags with at least 200 posts that the Danbooru Tag Taxonomy doesn't already sort.
AI classification: each tag was classified with Jev, TypeSafe's classification model, using the same slot definitions as the Danbooru tags.
Rules: tags like X_(cosplay) (dressed as character X) are clothing:outfit by rule, without asking the AI.

The English phrases for the approved tags are in another dataset, Booru Tags to Natural Language, together with the Danbooru ones.

What's inside?

Approved by category: appearance 3,254, pose 3,198, clothing 1,060, scenery 249, creature 206, equipment 128, other 96.
Post counts: straight from Rule34's API, from late September and October 2026. Only 12 tags have none, because Rule34 didn't return a count for them.

Limitations

Wrong slots. Expect some, mostly among rare tags.
needs_review tags have no usable slot. Only use them if you plan to sort them yourself.
One slot per tag, same as the Danbooru taxonomy.
Post counts are not live.
Rule34-only tags. Tags that Danbooru also has are in the Danbooru Tag Taxonomy instead.

What's not included

Artist, character, series and metadata tags aren't included.

Content warning: Rule34 is an adult site, and most of these tags describe nudity and sexual content. They're labelled like any other tag. The dataset contains tag names only, no images.

How to use it

If you don't code, the dataset includes a CSV file you can open in Excel or Google Sheets and filter by slot.

If you do, it's a couple of lines.

from datasets import load_dataset

ds = load_dataset("Mexes/rule34-tag-taxonomy", split="train")
approved = ds.filter(lambda r: r["status"] == "approved")

Releases are tagged, so you can pin a version with revision="v1.0".

The license is CC BY 4.0. You can use it for anything, including commercial projects, as long as you give credit. The tag names and post counts come from Rule34, so please credit them too.

FAQ

Found a tag in the wrong slot?

Leave a comment with the tag and where it belongs, since corrections are what this dataset needs most.

Why not add these tags to the Danbooru Tag Taxonomy?

That one has been checked much more carefully, and mixing in these tags would make it less reliable.

Built something with it?

Please share it in the comments. I'd love to see it.

Want to support me?

Thank you so much! You can Buy Me a Coffee, or send Buzz. It also helps a lot if you leave a comment and a rating.

Other articles that may interest you:

That's it!

Stay hydrated and don't forget to blink.