Get Chars Above Threshold

Scoped down to one category: this runs the same WD-style Danbooru tagger as the rest of the family, but only returns tags the model recognizes as character tags - hatsune_miku, 2b_(nier_automata), and the like - above your confidence threshold. Everything else (pose, clothing, setting) is discarded.

Danbooru-trained models like Illustrious and NoobAI learned enough named characters from their training data that, for anything with roughly a couple hundred works on the board, the character tag alone reliably resolves the look - no LoRA required. This node is the tool for finding that tag automatically: sort a mixed dataset by which character shows up in each image, flag images the tagger can't confidently attribute to any known character (useful for catching OCs, obscure characters, or mistagged data), or just confirm a generation actually produced who you asked for.

How it works

model_name (default EVA02_Large) picks the backbone - the same WD-tagger family used throughout this pack's tagger nodes: ViT_Large, SwinV2, ConvNext/ConvNextV2, MOAT, ViT, and _v3 revisions of a few of those. threshold (default 0.4) sets how confident the model needs to be before a character tag counts; character recognition tends to be noisier than general tags, so don't be surprised if you need to nudge this up for cleaner results on characters the model has seen less of. replace, matching the pattern across this whole tagger family, is most likely the underscore-to-space toggle for the raw tag text.

Inputs and outputs

• image - the IMAGE to check.
• threshold (default 0.4) - confidence floor for a character tag to be included.
• model_name (default EVA02_Large) - the tagger backbone.
• replace (default false) - likely swaps underscores for spaces in the output.

Output is a single STRING - character tags only, empty if the model doesn't recognize anyone in the image with enough confidence.

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. This is one of the pack's genuinely ML-heavy nodes - expect a model download the first time you pick a given model_name. If it fails on first run with an import error, check LogicUtils' auto-install hook (COMFYUI_LOGICUTILS_AUTO_INSTALL=1) or its requirements.txt before assuming something's broken.

Worth knowing

There's a second node in this pack, GetCharactersAboveThresholdFromTextNode, whose name suggests it should take a text input instead of an image - but its schema wants the exact same IMAGE input as this one. That's a real inconsistency in the pack, not documented anywhere; if you're expecting a text-in variant, you won't find one here. This is aria1th's (AngelBottomless's) LogicUtils pack, from the same handle that trains Illustrious XL - the connection is presumably why a full Danbooru tagger family exists inside what's otherwise a loose logic-node grab-bag. Public commits stopped in January 2026, and the README never got past "documentation is being prepared," so treat naming as a rough guide rather than a guarantee.