Comfyui-Anima-Tools: visual selectors for anime artists, characters and LoRA inside ComfyUI

What the eight nodes cover

The README puts eight core nodes under the AnimaArt category. Four are visual selectors: one for artist styles, one for anime characters, one for clothing tags, and a LoRA panel that searches, previews, downloads, favourites and loads. A fifth, the Anima Prompt Composer, randomises a whole prompt at run time. The rest are the Plus variants of the selectors and their supporting loader nodes.

The draw is that these are visual instead of typed. Rather than remembering a Danbooru artist spelling, you page through 3:4 preview cards and click, and the node fills itself in.

Artist tags and the @ normalisation rule

The artist library holds 40,000 plus Danbooru artist styles, sourced from the Anima-Style-Explorer project, with fuzzy name search and sorting by works count, uniqueness score, or alphabetically in either direction.

The detail worth knowing is the normalisation. In both the frontend and the Python node a selected artist is formatted as a tag with an @ prefix, so picking dairi yields @dairi, and stray by prefixes or duplicate symbols are stripped. That matters because an artist tag only works if it matches the token the model was trained on, and a hand-typed variant will not match.

The selector ships in two versions. The base Selector takes artist_tags, a mode combo of append or override, and an optional opt_prompt input, where append adds the artists to an existing prompt and override replaces the output. Selector+ adds extra_text and separator and cleans and de-duplicates when joining, so you do not get double commas or trailing spaces.

Character selector and the random composer

The character panel draws on AnimaDex and filters across gender, hair colour, eye colour, fan-art popularity measured as total illustrations, and source series. Selected names are cleaned of stray @ characters before output.

The composer is the more interesting node. At run time it picks randomly from the full artist, character, clothing, background and pose data, excluding user-defined items, and emits one string in a fixed order: artist, character, clothing, background, pose, with any category individually disableable. Characters support trigger, trigger_tags and trigger_random_tags modes. The random combination mode strips gender words such as 1boy and 1girl plus hair and eye colour words before sampling, then re-attaches hair and eye colour after the trigger word, keeping one colour per attribute and carrying no gender, which is what makes gender-swap prompts possible.

Clothing can come from the built-in library, a Danbooru tag group, or another character tags. The Danbooru path picks one main garment from tops plus bottoms, uniforms or traditional dress, then layers accessories, socks and shoes so the outfit does not contradict itself. Separate seeds exist for character and clothing, so you can lock one and keep rerolling the other.

Installing it in ComfyUI

Through ComfyUI Manager: open Manager, choose Custom Nodes Manager, search for Anima Tools, install, then restart ComfyUI.

Or by hand:

cd ComfyUI/custom_nodes/
git clone https://github.com/nregret/Comfyui-Anima-Tools.git

Restart ComfyUI and it loads. The README states a zero-dependency guarantee: the pack is plain frontend JavaScript plus native Python with no pip install of third-party libraries, and pyproject.toml confirms an empty dependency list requiring Python 3.9 or newer.

The LoRA safety rules in 3.3.1

pyproject.toml pins version 3.3.1, and the README leads with its LoRA rules. Loading and downloading accept .safetensors only, and files must sit in directories ComfyUI has already registered as LoRA directories. To use another disk or directory you register it under loras in extra_model_paths.yaml and restart ComfyUI first, after which the settings panel can select those directories and their existing subdirectories.

Credentials move in one direction. A saved Civitai key is never returned to the browser, so leaving the masked field unchanged keeps the key and clearing the field and saving removes it. Download redirects do not forward credentials, and remote LoRA previews only accept Civitai image URLs.

That is a tighter posture than most ComfyUI LoRA helpers take, and it is the first thing to re-check after any future update.

Frontend detail: 3:4 cards, pagination and CDN switching

The preview cards are 3:4, the ratio the README argues suits anime portraits, character art and CG, with a hover zoom and shadow. The pagination control shows the current page and total records and includes a jump box, where typing a page number and pressing enter goes straight there.

Sample images load through one of several CDN routes, JsDelivr, GitHub Raw and Statically, switchable from the panel. That is a practical concession to the fact that a visual selector is useless when the thumbnails do not load.

The interface is bilingual and follows the ComfyUI native language setting, Comfy.Locale, automatically, with no separate configuration.

Where the data comes from, and maintenance

Artist data comes from Anima-Style-Explorer and character data from AnimaDex, both credited in the README, so the pack is an interface over other people datasets rather than a source of truth. The last push was on 2026-09-15, and pyproject.toml declares MIT with version 3.3.1.

The obvious limit is scope. This is built for anime diffusion workflows and Danbooru-shaped tag conventions, so the @ artist prefix and the character trigger logic both assume a model trained that way. Point it at a photorealistic checkpoint and the selectors will emit tags that mean nothing to it.

Editorial conclusion

Install Comfyui-Anima-Tools if you build anime prompts in ComfyUI and would rather click through 40,000 artist styles than recall spellings, or if you want a composer that holds a character identity steady while rerolling outfits. Skip it for photorealistic checkpoints, since the Danbooru tag conventions will not transfer. Register your LoRA directories in extra_model_paths.yaml before you open the LoRA panel, because 3.3.1 refuses to load .safetensors files from anywhere ComfyUI has not already registered.

Frequently asked questions

Official sources

1. Issues
2. nregret/Comfyui-Anima-Tools on GitHub
3. README