Get All Tags Above Threshold Except Characters

Same Danbooru-style auto-tagger as GetAllTagsAboveThresholdNode, with one deliberate omission: whatever the model recognizes as a character tag gets left out of the result. You get pose, clothing, setting, expression, all the general descriptive tags - just not "this is Hatsune Miku."

That's a real, specific need in Danbooru-vocabulary captioning: sometimes you want a dataset caption to describe what's happening in an image without baking in a character identity, because you're training a pose or style concept you want to generalize across characters rather than lock to one. Tag a batch of reference images with the character name still attached, and a LoRA can end up quietly tying a pose or outfit to that one character instead of learning it as a portable concept. Stripping character tags at capture time - rather than filtering them out after the fact - is the cleaner way to avoid that.

How it works

Same mechanism as its sibling: a WD-style tagger backbone (model_name, defaulting to EVA02_Large, with the same set of architecture choices - ViT_Large, SwinV2, ConvNext/ConvNextV2, MOAT, and their _v3 revisions) runs over the image and returns every tag above threshold, minus anything the model's character category catches. replace, following the convention shared across this whole tagger family, most likely toggles whether underscores in raw tags become spaces in the output.

Inputs and outputs

• image - the IMAGE to tag.
• threshold (default 0.4) - confidence cutoff for a tag to be included.
• model_name (default EVA02_Large) - the tagger backbone.
• replace (default false) - likely the underscore-to-space toggle.

Output is a single STRING - the tag list, characters excluded.

Installing ComfyUI-LogicUtils

ComfyUI Manager: Install Custom Nodes → search "ComfyUI-LogicUtils" → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/aria1th/ComfyUI-LogicUtils

Restart ComfyUI. Expect a weights download the first time you use a given model_name - this is a real ML model, not a lightweight logic node. If it errors on first run with a missing package, check the pack's auto-install hook (COMFYUI_LOGICUTILS_AUTO_INSTALL=1) or its requirements.txt.

Worth knowing

If you specifically need character identity in the output - say, sorting a mixed dataset by which character appears - reach for GetCharactersAboveThresholdNode instead, which is the mirror image of this node: characters only, nothing else. This is aria1th's (AngelBottomless's) LogicUtils pack; the same person trains Illustrious XL, which is presumably why a Danbooru-style tagger family shows up at all in an otherwise unrelated grab-bag of logic nodes. The README doesn't document the exact tag-category logic here, and public commits stopped in January 2026, so the "except character" behavior is read directly off the node's name rather than confirmed against the source.