Create Danbooru and e621 Tag Lists for AI related Autocomplete Extensions

This script scrapes tags from the Danbooru API and downloads e621 tags from https://e621.net/db_export/ and formatting them for AI autocomplete extensions. It can create separate and merged tag lists for Danbooru and e621, with options like alias inclusion, minimum post thresholds, and filtering by alias status as well as creating Krita AI diffusion compatible tag lists (unfiltered/NSFW versions only).

The CivitAI page for these can be found here: https://civitai.com/models/950325 An archive of CSV files I made can be found here: https://github.com/DraconicDragon/dbr-e621-lists-archive

The tag lists are automatically created (as GH actions artifacts) every month in the archive repo, and saved every 3 months in the tag-lists folder.

Note on Gelbooru option: The script also supports scraping from Gelbooru. Don't use it. Because of how their API works, or at least how I know it works, the script will have to go through every single page of tags to get the full data to then be able to create the tag lists - which will take 4-5 hours. Here's the data for it https://huggingface.co/datasets/DraconicDragon/Gelbooru-Tags-Full_2026-06-11 and here's a tag list with default settings (and post_threshold = 20) https://github.com/DraconicDragon/dbr-e621-lists-archive/blob/main/tag-lists/gelbooru/gelbooru_2026-06-10_pt20.csv There's also no aliases

Danbooru's Google BigQuery dump notes, most can just ignore this unless interested

If you dont want to scrape danbooru's API using this script and know your way around Google BigQuery, there's an official danbooru dump [here](https://console.cloud.google.com/bigquery?project=danbooru1&pli=1) where you can easily do SQL queries. Krita-AI-diffusion has a guide-ish file [here](https://github.com/Acly/krita-ai-diffusion/tree/main/ai_diffusion/tags) that explains some stuff, maybe it helps

If you are a developer

... looking to use these in some way, please don't rely on the merged list - it is not very future proof or scalable. Instead I recommend and hope that you will implement a way to load multiple CSV files in your project. The merging method used by my script came to existence purely because of convenience - it worked with existing autocomplete extensions with no or only minor issues (which didnt impact functionality) and did not require third parties to write extra code for it.

Extra notes about how bad the current merging method is

The current merging method simply increases E621's category count by 7, which is in danboru's invalid region (above 5/6), so they don't collide with each other. This is not future proof because if danbooru decides to add 2-3 more categories, they will collide - updating the merged list would then require an update to the autocomplete extensions using the merged list. The same thing is about scalability - If there is a new service that also has a similar structure/tags (maybe gelbooru _{gelbooru's api responses are garbaw, but i havent dug further into it}, rule34 or bbooru?) then how would that be added to the merged list? Yes categories can just be increased again, but this is all hardcoded and shouldn't be. I did think of a way like adding a "service" column that has a string like "danbooru,e621" or just "danbooru" and "e621" (depending on tag, more robust if category numbers are different and same numbers mean different things) but at that point writing the code to support this would be just as much as supporting multiple files loaded (probably not but im also kind of trying to convince whichever dev might be reading this). And also at that point having a single file to load for tags the user probably doesnt want anyway might be suboptimal either way.

TLDR: Just don't write your code around the merged list and in best case support loading multiple CSV files.

Links for programs and their autocomplete extensions

The named programs and their extensions below are only the ones I have tested and used myself. The default tag lists will likely work with other autocomplete programs/extensions.

• A1111/Forge/reForge:

  • Extension: DominikDoom/a1111-sd-webui-tagcomplete
  • Location of the tag lists should be in /webui/extensions/a1111-sd-webui-tagcomplete/tags/
  • To select the file you want to use go to your WebUI's settings tab, search for "tag filename" or look for "Tag Autocomplete" in the left sidebar and click that to get the tagcomplete settings and then select the tag file from the "Tag filename" drop-down.

• ComfyUI:

  • Custom Node 1: pythongosssss/ComfyUI-Custom-Scripts

    • ComfyUI Settings -> pysssss -> Manage custom words
      • I recommend pasting the CSV file content in the big textbox and saving it or using a link like this one and load it. (gotten from here by pressing "Raw" button at top right)
      • When you press save it might take a short bit and the "Saved!" appears only for less than a second. I'm unsure if you need to wait for it to show to be able to dismiss the window/popup.

  • Custom Node 2: jupo-ai/comfy-ex-tagcomplete

    • Seems like this extension already comes with tag lists that seems to be from the a1111-sd-tagcomplete extension, so you might not need to add any custom ones.
    • To add custom CSV files: Drop the CSV file you want to use in the ComfyUI\custom_nodes\comfy-ex-tagcomplete\tags
      • Refresh the ComfyUI tab if one is open, and then check ComfyUI Settings -> jupo -> tags file

• SwarmUI:

  • Please see: Autocompletions.md#word-lists

• Stability Matrix: Settings -> Inference -> Auto Completion

• Krita AI diffusion:

  • Go to Settings -> Interface -> Tag Auto-Completion
    • You will see a refresh and folder icon on the right side. Click the folder icon and it should take you to the tags folder where you have to replace the tag files with the ones you want to use. _{Note: For me personally on Windows, the tags folder was in %appdata%\krita\pykrita\ai_diffusion\tags while the folder icon took me to %appdata%\krita\ai_diffusion\tags}

Running the Script Yourself

• Clone the repository git clone https://github.com/DraconicDragon/danbooru-e621-tag-list-processor
• cd danbooru-e621-tag-list-processor into the cloned repository folder
• Have Python installed and create a venv python3 -m venv .venv
• Install the dependencies pip install -r requirements.txt
• Run python main.py and fill out the options - after this the processing begins
  • You can also just spam enter to use the default values which will give you both Danbooru and e621 tag lists including active and deleted aliases and the merged list.

The tag lists should be saved in a folder called /output/tag_lists that will be created after the program successfully creates the CSV files (to check you can just check the terminal output, it should print out the location of the saved files)

Raw mode

There is a "Raw mode" which will allow you to scrape the raw data without turning it into tag lists. Simply enter "raw" for the first question/selection and follow the instructions. Don't use with e621 because code no work for that, just download from here is easier anyway: https://e621.net/db_exports

Danbooru scraping part in the code was originally copied from here (many thank): https://github.com/BetaDoggo/danbooru-tag-list

• interesting read on making somewhat separated SFW/NSFW tag lists https://github.com/Acly/krita-ai-diffusion/tree/main/ai_diffusion/tags