This character tagger is the same AI anime tagger as our Danbooru tagger, with the character results moved to the top. It knows 2,132 characters by name, from Hatsune Miku and Reimu Hakurei to Ganyu and Mash Kyrielight, and shows how sure it is about each one. Below you can also search every name it knows, which is the quickest way to check whether your favorite is in there.
How it recognizes characters
The tagger is the WD14 ConvNeXt V2 model (the "WD tagger"). It learned by studying millions of pictures from Danbooru, an anime art site where volunteers tag every image, including who is in it. A character made it into the model only if they had at least 600 tagged pictures on the site, so the list is made up of characters with a lot of fan art.
Two things follow from that:
- Popular characters work best. The more fan art a character has, the more examples the model saw, from different artists, outfits and angles.
- Newer characters are missing. The training pictures stop at the end of 2022, so characters who became popular after that aren't in the list, no matter how famous they are now.
Reading the results
Each character comes with a percentage, which is how confident the tagger is. By default it only names a character at 85% or more, because a confident wrong answer is worse than no answer. You can change that with the "How sure must it be?" slider.
- A name with a high score (90% and up) is usually right, especially for well-known characters in their usual outfit.
- "Closest guesses" appear when nothing reaches the cut-off. Treat them as hints, not answers. They're often a character with a similar look or from the same series.
- Two or more names can appear when a picture shows several characters, or when the tagger can't decide between two lookalikes.
- The series name in brackets, as in
ganyu_(genshin_impact), is part of the official tag. Danbooru adds it when a name could mean more than one character.
The regular tags underneath (hair color, outfit, pose) are worth a look too. They often explain a miss: if the tagger sees swimsuit on a character who is usually in armor, it may not connect the two.
When a character isn't recognized
If you get no name, run through this list:
- Search the list below. If the character isn't there, the tagger can't name them. That usually means fewer than 600 tagged pictures by the end of 2022.
- Crop to the character. Every picture is shrunk to 448 × 448 pixels before tagging. A small character in a big scene may only be a few dozen pixels tall by then.
- Try a more typical picture. Unusual outfits, heavy stylization, chibi versions and partial views (just a hand, or the back of the head) are much harder.
- Lower the slider a little. A real match sometimes sits at 70 to 80%. Check the result against what you know before trusting it.
This tool names drawn characters. It isn't built to identify real people, and you shouldn't use it to try. Photos of cosplayers may get the character's tag if the costume is very distinctive, but results on photos are unreliable.
Search all 2,132 characters
Type part of a name, in any order of first and last name, or a series such as kancolle or genshin. Names use Danbooru's format: lowercase, with underscores between words and Japanese names family name first (hakurei_reimu, not "Reimu Hakurei").
Loading the character list…
The 50 most common characters
These are the characters with the most tagged pictures in the model's training data, so they're the ones it recognizes most reliably. The counts are Danbooru posts with that character tag when the model's tag list was built.
| # | Character tag | Tagged pictures |
|---|---|---|
| 1 | hatsune_miku | 78,616 |
| 2 | hakurei_reimu | 67,710 |
| 3 | kirisame_marisa | 62,327 |
| 4 | remilia_scarlet | 46,894 |
| 5 | flandre_scarlet | 43,434 |
| 6 | izayoi_sakuya | 42,465 |
| 7 | admiral_(kancolle) | 34,812 |
| 8 | artoria_pendragon_(fate) | 33,041 |
| 9 | alice_margatroid | 32,516 |
| 10 | kochiya_sanae | 32,504 |
| 11 | patchouli_knowledge | 32,120 |
| 12 | konpaku_youmu | 31,146 |
| 13 | cirno | 30,904 |
| 14 | yakumo_yukari | 29,607 |
| 15 | komeiji_koishi | 27,406 |
| 16 | shameimaru_aya | 25,978 |
| 17 | fujiwara_no_mokou | 24,550 |
| 18 | reisen_udongein_inaba | 24,375 |
| 19 | hong_meiling | 23,457 |
| 20 | akemi_homura | 23,123 |
| 21 | komeiji_satori | 22,757 |
| 22 | kaname_madoka | 22,080 |
| 23 | saigyouji_yuyuko | 21,758 |
| 24 | kaga_(kancolle) | 20,549 |
| 25 | inubashiri_momiji | 20,497 |
| 26 | yakumo_ran | 18,043 |
| 27 | kagamine_rin | 17,416 |
| 28 | konpaku_youmu_(ghost) | 17,065 |
| 29 | moriya_suwako | 17,047 |
| 30 | rumia | 16,841 |
| 31 | miki_sayaka | 16,686 |
| 32 | kaenbyou_rin | 16,675 |
| 33 | kazami_yuuka | 16,551 |
| 34 | shimakaze_(kancolle) | 16,285 |
| 35 | reiuji_utsuho | 16,263 |
| 36 | saber | 15,967 |
| 37 | hibiki_(kancolle) | 15,954 |
| 38 | chen | 15,887 |
| 39 | kamishirasawa_keine | 15,388 |
| 40 | tatara_kogasa | 15,022 |
| 41 | kawashiro_nitori | 14,973 |
| 42 | mash_kyrielight | 14,919 |
| 43 | hinanawi_tenshi | 14,700 |
| 44 | sakura_kyouko | 14,333 |
| 45 | tomoe_mami | 14,245 |
| 46 | shigure_(kancolle) | 14,238 |
| 47 | houraisan_kaguya | 13,824 |
| 48 | koakuma | 13,641 |
| 49 | kongou_(kancolle) | 13,106 |
| 50 | ganyu_(genshin_impact) | 12,981 |
Touhou Project dominates the top of the list: 16 of the top 20 are Touhou characters, thanks to that series' huge fan-art community. The other four are Hatsune Miku (Vocaloid), the Admiral from Kantai Collection, Artoria Pendragon (Fate) and Akemi Homura (Madoka Magica). Kantai Collection has the most characters overall, with 234 on the list, and Genshin Impact has 45. Further down, the counts drop quickly: half of all 2,132 characters have fewer than 1,362 tagged pictures.
Using character tags in prompts and captions
Press Copy tags to get the character name along with the rest of the tags. If you're pasting into an image generator such as AUTOMATIC1111 or Forge, keep "Prompt-ready brackets" ticked, so ganyu (genshin impact) becomes ganyu \(genshin impact\) and the brackets aren't read as emphasis. If you're captioning a folder of the same character for a LoRA, use the batch tagger and read our LoRA captioning guide first. It explains when to keep the character tag and when to replace it with your own trigger word.
Frequently asked questions
Is this a "waifu tagger"?
People sometimes search for a waifu tagger when they want to know who their favorite anime character is, or to tag a collection of character art. That is exactly what this page does: it names anime and game characters and tags everything else in the picture, whoever your favorite happens to be.
How accurate is it?
For popular characters in their usual look, high-confidence answers are usually right. Accuracy drops for rarer characters, unusual outfits, small or partial views and very stylized art. The model card publishes one overall accuracy figure for all tags (an F1 score of about 0.69 at its best threshold), not a separate figure for characters, so treat every answer as a strong hint and double-check anything that matters.
Why is a really famous character missing?
Either they became popular after 2022, when the training data ends, or they had fewer than 600 tagged pictures on Danbooru at that point. Search the list above to check. Newer tagger models exist and know more recent characters; we may add them in the future.
Does my picture get uploaded?
No. The character tagger runs in your browser after a one-time model download, so your picture never leaves your device. How in-browser AI works shows how to check this yourself.