Datasets:

DraconicDragon
/

Gelbooru-Tags-Full_2026-06-11

tag_id int64 1 1.9M	tag_name stringlengths 0 255	post_count int64 0 12M	category_id int64 0 6	is_ambiguous bool 2 classes
1	asahina_mikuru	5,107	4	false
2	cosplay	221,731	0	false
3	suzumiya_haruhi_no_yuuutsu	28,239	3	false
4	waitress	18,630	0	false
5	aizawa_yuuichi	185	4	false
6	crossover	96,399	0	true
7	gundam	82,991	3	true
8	gundam_seed	8,644	3	false
9	minase_akiko	631	4	false
10	parody	95,853	0	true
11	spoilers	41,427	5	false
12	what	11,221	0	false
13	seifuku	104	6	false
14	suzumiya_haruhi	12,913	4	false
15	misteor	694	1	false
16	reborn	99	4	false
17	harada_takehito	1,293	1	false
18	pleinair	534	4	false
19	bdsm	191,391	0	false
20	bondage	152,178	0	true
21	femdom	44,437	0	true
22	riding_crop	5,920	0	false
23	thighhighs	1,649,967	0	false
24	tongue	683,081	0	false
25	game_cover	361	0	false
26	japanese	214	6	false
27	breasts	5,898,076	0	false
28	cum	672,458	0	false
29	eyeshield_21	681	3	false
30	kiss	110,878	0	false
31	paizuri	115,985	0	false
33	funny	329	6	true
34	panties	1,017,419	0	true
35	panty_pull	80,050	0	false
36	peace@pieces	180	3	false
37	male	676	6	false
38	yaoi	173,687	0	true
39	shota	145,903	0	false
40	pico	22	6	false
41	comic	240,095	0	false
42	water	303,575	0	false
43	yellow	174	6	false
44	cold	4,564	0	false
45	ribbons	766	0	false
46	wings	618,208	0	false
47	blue_hair	1,298,595	0	true
48	feathers	97,925	0	false
49	kotonomiya_yuki	438	4	false
50	nude	1,186,329	0	false
51	red_eyes	1,836,072	0	true
52	short_hair	3,211,049	0	false
53	suigetsu	897	3	false
54	buruma	45,072	0	false
55	higurashi_no_naku_koro_ni	14,540	3	false
56	monochrome	654,981	0	false
57	shichimenchou	1,064	1	false
58	sonozaki_mion	2,608	4	false
59	sonozaki_shion	1,937	4	false
60	yuri	323,366	0	true
61	aria_(sister_princess)	163	4	false
62	innertube	40,690	0	false
63	school_swimsuit	74,998	0	false
64	sister_princess	1,263	3	false
65	purple_hair	964,149	0	true
66	black_hair	2,373,017	0	true
67	blue_eyes	2,693,065	0	false
68	bra	322,645	0	false
69	game_cg	302,132	6	false
70	hirose_misaki	4	4	false
71	lingerie	89,921	0	true
72	nurse	29,633	0	false
74	:3	156,418	0	false
75	obey	23	0	false
77	4koma	67,360	0	false
78	chibi	322,583	0	false
80	suicide	1,315	0	false
81	akizuki_ritsuko	3,363	4	false
82	antenna_hair	112,048	0	false
83	camisole	60,605	0	false
84	glasses	538,324	0	false
85	hat	1,603,655	0	false
86	idolmaster	297,399	3	true
87	wet	230,567	0	false
88	wink	113,542	0	false
89	guilty_gear	28,217	3	false
90	kuradoberi_jam	1,168	4	false
91	bikini	763,849	0	false
92	blush	4,625,612	0	false
93	mole	387,353	0	false
94	pink_hair	1,092,055	0	false
95	redhead	277	6	false
96	tokimeki_memorial_2	664	3	false
97	amami_haruka	6,384	4	false
98	gloves	1,949,310	0	true
99	microphone	69,532	0	false
100	amamiya_manami	124	4	false
103	massugu_go	31	0	false
104	kantoku	2,591	1	false
105	miniskirt	359,783	0	false
106	odori_momoha	74	4	false

End of preview. Expand in Data Studio

Tags fetched from Gelbooru's API on 11th June 2026. The API is ass, by the way.

Update 16.06.2026 > Thanks to a user an issue where special characters were escaped (eg: ' for apostrophe) was caught and fixed. If you scrape Gelbooru's tags then beware of this, since the API returns them with this.

The data for both the .parquet and JSONL file are the same*
And I believe it is all the tags - means you have some pruning to do if you are going to use this downstream. Maybe start with category 6 deprecated tags

What the data includes: (Basically anything that the tags endpoint would give me)

Tag ID
Tag Name
Post Count
Category ID (All categories 0 to 6; you can find a table at the bottom with names for the IDs)
If tag is ambiguous

It seems like there's an empty tag/string that is applied to almost all posts (1.6m posts don't have it) and thus has the highest post count number
The .parquet file is the one you should probably use.
The JSONL version is closer to Gelbooru's API response with (I believe) only difference being that it's newline delimited instead of an array of items.

There's some other rather trivial differences between the Parquet and JSONL versions:

Sorting:

Parquet file is sorted by tag id (ascending); You can just use the HF Dataset Viewer to sort it by different columns
JSONL file is sorted by (post) count

Original Key (JSONL)	Renamed Key (Parquet)
`id`	`tag_id`
`name`	`tag_name`
`count`	`post_count`
`type`	`category_id`
`ambiguous` (0 or 1)	`is_ambiguous` (bool)

Why?
I could've just not uploaded the JSONL but I thought it to be potentially useful for more casual human viewability and API compatibility

Category ID -> Category name table if you need:

ID	Category Name
0	General
1	Artist
2	Invalid / Unused
3	Copyright
4	Character
5	Meta
6	Deprecated

Code used: https://github.com/DraconicDragon/danbooru-e621-tag-list-processor/blob/main/tag_lists/gelbooru.py
Took about 4-5 hours. I could not find a way to sort the tags by count, so the only way I saw was fetching ALL the tags - and that is how it came to this repo.
The response frequency slowed down by a decent amount after 50% of all the pages were done (max page ID/PID as of making this was around 13935 with limit=100)

Downloads last month: 490