Danbot Load Ban Tags

Danbot Load Ban Tags is the pack's answer to "the model keeps putting in tags I don't want." You connect its output to the ban_tags input on the Danbot Generator or V2408 Pipeline, and those tags get blocked during generation - mechanically enforced, not just suggested in a negative prompt.

There's a real distinction worth understanding: the Danbot model doesn't have a negative prompt box. The way you steer what it doesn't emit is through banned tokens, which the generator turns into bad_words_ids - the model literally can't generate a sequence containing those tokens. That's a much harder constraint than image negative prompts, which are advisory. It's also why this is a small utility node and not a settings toggle: the ban list is data, and this node is how you load data.

What it does

One input, one output. The input is template_name, a dropdown listing the ban files shipped in the pack's tags/ban_template directory:

• all_text.txt - the kitchen sink: character names, holiday tags, and other stuff you don't want creeping into a tag list.
• major_concepts.txt - broad high-level concepts.
• year.txt - year tags (handy: a model with a 2024 cutoff shouldn't emit "2026").
• military.txt - military tags.
• alternate_attire.txt - a tiny list of outfit-related tags.

The output is ban_tags, a single comma-separated string. Leave the dropdown empty and you get an empty ban list - a no-op, which is fine if you just want the wire in place.

The part people misread

The ban operates on the tag model's output tokens, not on the final image prompt. Banned tags are stripped from the Danbooru tag list before it ever reaches your text encoder. If you ban school uniform, you're not putting anything in the image negative prompt - you're making sure the translation model never suggests the tag in the first place. Different layer, different effect. If you wanted both, wire the ban list and a normal negative prompt.

One implementation detail that can bite: banned tags are matched against the model's tokenizer vocabulary, so an exact tag that isn't in the vocab just silently does nothing. The loader supports * wildcards (e.g. *202* to block any year) - use those when you're not sure the exact tag exists.

Install

Same as the pack - ComfyUI Manager → danbot-comfy-node, or:

cd ComfyUI/custom_nodes
git clone https://github.com/p1atdev/danbot-comfy-node
cd danbot-comfy-node
pip install -r requirements.txt

(portable: ../../../python_embeded/python.exe -s -m pip install -r requirements.txt.) Restart ComfyUI and it's under prompt/Danbooru Tags Translator. You can also drop your own .txt files into the pack's tags/ban_template/ folder - newline-separated tags, // for comments - and they'll appear in the dropdown after a restart.