Autocomplete tag packs
ComfyAssets prepares filtered autocomplete data from Danbooru's public
tags.json and tag_aliases.json APIs. Dataset releases are versioned separately
from KikoTools and stored as release assets, keeping bulk data out of Git history.
Danbooru — Illustrious / NoobAI, 2026-10-16
Download the CSV and manifest. This snapshot contains 165,158 canonical tags, their categories, post counts, and aliases. The adjacent manifest records the retrieval date, model targets, selection filters, schema, byte size, and SHA-256 checksum.
The UTF-8 CSV has no header. Columns are tag,category,post_count,aliases.
Aliases share a comma-separated field; use a CSV parser that respects quoting.
An empty alias field means no aliases. Category codes are 0 (general), 1
(artist), 3 (copyright), 4 (character), and 5 (meta).
In ComfyUI Settings, search Embedding Autocomplete, open its settings, then download the pack and choose the CSV under Tag list file. Enable Show Custom Words and select Save Settings. Importing stages the list; Cancel leaves the active list unchanged. The browser caches the saved file in IndexedDB for offline use and restores it after a refresh. Clearing browser site data removes this cache. There is no server-side autocomplete folder to populate.
Alternatively, enter a CSV/TXT URL and select Load URL, then save. Remote
servers must allow browser cross-origin requests; for GitHub release assets,
download and import the file locally. Existing one-column word lists and
tag,frequency lists remain supported. The default URL is unchanged.
Parsing runs in a Web Worker with a 32 MB UTF-8 limit. Search checks canonical names and aliases, inserts the canonical tag, and ranks exact matches ahead of prefixes and substrings, then by descending post count. The number of displayed results is bounded by Maximum suggestions. Categories are retained in records; category-specific controls are not currently exposed.
Source and terms
Source data: Danbooru, collected 2026-10-16. Selection and packaging: ComfyAssets. This is an independent filtered snapshot, not an official Danbooru release or a guarantee of model compatibility.
Refer to Danbooru's Terms of Service and the official terms template. The template's Copyright Infringement section identifies tags as factual information that is generally not copyrightable. Its API Usage section requires compliance with API rate limits. This reference does not claim an explicit dataset license or extend KikoTools' software license to upstream data.