Ban Tags 🍌

This node only matters if you're using cgem156's Dart nodes - the part of the pack built around p1atdev/dart-v1-sft, a small model trained specifically to generate Danbooru-style tag strings, essentially autocompleting a plausible set of tags the way a Danbooru-trained checkpoint would want them. If that's not part of your workflow, skip this article; Ban Tags has no purpose on its own.

What it's for

Tag-generator models like Dart will happily produce tags you don't want in a given run - a character tag when you wanted something generic, a rating tag that doesn't match your intent, a concept tag that keeps showing up uninvited. Ban Tags takes a list of tags you want excluded and, using the same tokenizer the Dart model uses, turns that list into a form the generator node can actually enforce against - a ban list scoped to the generator's own vocabulary rather than a plain-text blocklist that might not match how the model tokenizes things.

Inputs and outputs that matter

• tokenizer - a DART_TOKENIZER, supplied by the pack's Dart loader node, not something you configure here.
• ban_tags - multiline text, the tags you want excluded from generation.
• Output: STRING - formatted for the Dart generation node to consume as its exclusion list.

Installing it

ComfyUI Manager → search cgem156-ComfyUI🍌 → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/laksjdjf/cgem156-ComfyUI

then restart ComfyUI. This node depends on the pack's Dart tokenizer/loader node upstream, which pulls p1atdev/dart-v1-sft from Hugging Face the first time you run it - expect that download before this part of the pack works at all.

Common issues & troubleshooting

No tokenizer input available. This is the single required piece that makes the node function - you need the pack's own Dart loader node in your graph first; a generic tokenizer from elsewhere won't plug in, since the type is specific to this node family.

Banned tags still showing up in generation. Double-check spelling and formatting matches how the tag actually appears in the Dart model's vocabulary - Danbooru tags are specific (underscores, exact wording), and a near-miss spelling won't match what you're trying to exclude.

Not sure this is even the node you need. If you're not already running Dart's tag-generation nodes elsewhere in your workflow, this one won't do anything useful in isolation - it's a modifier for that specific sub-pipeline, not a general-purpose "remove tags from my prompt" tool. If that's what you're after, you want simple text editing or a dedicated prompt-filtering node instead.