CharaDB: 1,027 free anime prompt materials (hair, outfits, poses + wildcards)

I make AI-generated CG collections (300–500 pages each), and the thing that eats my time is not "getting one cute image". It is finding the words for a specific outfit, and keeping the same character for 300 pages after the outfit changes.

So I built a database and I'm giving it away.

CharaDB — https://charadb-2fi.pages.dev/

• 1,027 images, every one with ready-to-paste Danbooru tags

• Hairstyles: 46 styles × 5 colors (brown / black / blonde / pink / silver), 12 bangs, 20 hair colors, 16 accessories, 19 eye colors, 9 pupil shapes

• Outfits: 440, in 15 categories (school uniforms, casual spring-summer / autumn-winter, office, dresses, kimono / yukata, beachwear, sports, jobs & cosplay, fantasy, loungewear, seasonal events, idol stage, gothic-lolita / jirai-kei, military / sci-fi)

• Poses: 281 (standing incl. side views, sitting, lying, crouching / kneeling, upper-body gestures, action, everyday actions, two-girl poses). Every pose also has a line-art version for ControlNet (lineart / canny).

• A single-file web app: pick a hairstyle, eyes, outfit and pose → choose expression / camera / background → copy the full prompt. Quality-tag presets for Illustrious / NoobAI, Anima, NovelAI and Pony. Negative prompt comes with it.

Why every image uses the same character

All 1,027 images were generated with the same checkpoint (JANKU v4, NoobAI-XL / Illustrious based), same sampler settings and the same fixed seed, on a fixed base character (brown hair, medium hair, blunt bangs, brown eyes). Only the outfit or pose changes between cards, so you can compare "what does shirt dress, striped shirt, long dress, belt actually look like" against the card next to it.

The outfit and pose cards contain only the outfit / pose tags — no character tags — so you can drop them straight into your own character prompt.

No reference photos and no product images were used; everything was generated from text in my own environment.

How to use the app

1. Open https://charadb-2fi.pages.dev/ (works on phone too)

2. In the Hair / Outfit / Pose tabs, press "add" on a card (hair cards have color dots for the 5 colors)

3. Press "Build" at the bottom, choose expression / camera / background

4. Copy prompt — paste into your generator

Click any card image to open the full-size version. On pose cards, toggle "line" to get the line art for ControlNet.

Using CharaDB as wildcards (Dynamic Prompts)

The tag column is also exported as plain wildcard files, one item per line:

charadb_hair.txt         46 hairstyles (no color — add your own)
charadb_hair_color.txt   20
charadb_bangs.txt        12
charadb_hair_acc.txt     16
charadb_eyes.txt         28 eye colors / pupil shapes
charadb_outfit.txt       440 outfits
charadb_pose.txt         281 poses

Put them in extensions/sd-dynamic-prompts/wildcards/ (A1111) or your ComfyUI wildcards folder, then:

masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest,
1girl, solo, __charadb_hair__, brown hair, __charadb_eyes__,
__charadb_outfit__, __charadb_pose__,
simple background, white background

Tip: lock the hair / eye lines once you like them, and only randomize __charadb_outfit__ and __charadb_pose__. That is the whole trick behind "same character for 300 pages": fixed lines stay byte-identical, you only change the lines you mean to change.

A template I use:

[quality]   one line for your model
[fixed]     hairstyle / bangs / hair color / eyes / body  ← never edit these
[variable]  outfit / pose / expression  ← change at most 3 per page
[scene]     background / camera
order: quality → 1girl, solo → fixed → expression → outfit → pose → camera → background

Download

• Web app (no download needed): https://charadb-2fi.pages.dev/

• Full zip with images, line art, app and wildcards: GitHub Release — https://github.com/taki-ttt/charadb/releases/tag/v1.0

Terms

• Free to use, no credit required.

• Please do not redistribute or resell the images themselves (original or edited).

• Not for depicting real people or minors.

• Generated with JANKU v4 (NoobAI-XL / Illustrious based). The base models' terms (NoobAI-XL 1.0 License / FAIPL-1.0-SD) also apply.

• Tags, prompts and the app may be used commercially. Treat the images as reference material (samples / ControlNet guides); do not include the images themselves in works you sell (per the base model's terms).

Extended edition

Extra outfits and poses and the bulk download of all 1,097 originals are not in the public edition. They are distributed through my LINE open chat (Japanese): https://line.me/ti/g2/OhUywV8y_UynwVUw9p0FbYAI9zhruNMIqs4qTg

Author: Taki — https://x.com/taki_auto (I sell AI-made CG collections on Japanese doujin platforms and teach the process.)