Remove Banned Tags From Tags [LP]

If you're pulling tags off an autotagger or a caption dataset and feeding them into a prompt, you'll eventually hit tags you never want showing up - watermark, signature, whatever junk crept into the training data's own captions. This node exists to filter those out before the tag string reaches your prompt.

What it does, and the thing that isn't in the node

Feed it a comma-separated tag string and it hands back a cleaned version. Here's the part worth being upfront about: there is no banned_tags field on this node. You don't get a text box to type your own blocklist into - the node's only input is the text you're filtering. That means whatever gets removed is coming from a list baked into the node's own code rather than something you configure per-run. The pack credits the underlying tag-filtering logic to comfyui_tag_fillter, so the blocklist is presumably whatever that library ships with by default. If you were expecting to customize what counts as "banned," this particular node doesn't give you that lever - that's the honest limit here, and it's worth testing on a sample tag string before you trust it in a real pipeline, since the exact contents of that list aren't documented anywhere accessible from the node itself.

This node's sibling, Remove Banned Tags From Text, does the same filtering job but is built for looser natural-language captions instead of a strict comma-joined tag list - reach for whichever matches the shape of the text you're actually holding.

Inputs and outputs that matter

One field, and it's the whole node:

• text (multiline, defaults to empty) - your comma-separated tags, going in.

The output is tags TEXT - the filtered string, ready to drop straight into a prompt or a downstream tag-processing node like Tag Category Filter or Tag Category Keeper from this same pack.

Installing it

Through ComfyUI Manager, search "ComfyUI Level Pixel" or "ComfyUI-LevelPixel." Or:

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

Restart ComfyUI. No models, no extra dependencies to install - it's a text-processing node.

Where people get burned

The main trap is expecting a configurable blocklist and not finding one - go in assuming this filters against a fixed, undocumented set of tags rather than anything you can edit from the node UI. If a specific tag you're trying to remove keeps surviving the filter, that's a strong signal it's not on whatever list is hardcoded in, and this node isn't the place to add it.

Format mismatch is the other common issue: if your input is prose rather than a clean comma-separated list - full sentences, natural-language captions - feed it to Remove Banned Tags From Text instead. This one is built around the tag-list shape (booru-style, one concept per comma), and results on freeform text are unlikely to be what you expect.