Get Rating Class

The same WD-style tagger backbone that powers the rest of this pack's tagger nodes carries a second output alongside its tag predictions: a content-rating classification. This node exposes just that - you get back a rating class as a string, following the Danbooru convention these models were trained on (Danbooru's own taxonomy runs general, sensitive, questionable, explicit; the training-era WD taggers commonly collapse this to something like general/sensitive/questionable/explicit or a three-tier safe/questionable/explicit depending on the model generation).

The obvious use is gating: branch a workflow so anything the classifier calls explicit gets routed differently - flagged, blocked, sent to a different output folder - before it goes anywhere public-facing. It's also useful for dataset triage: sorting a big batch of images by content rating without opening every single one.

How it works

model_name (default EVA02_Large) selects which backbone runs the classification, from the same option set as this pack's other taggers - ViT_Large, SwinV2, ConvNext/ConvNextV2, MOAT, ViT, and a few _v3 revisions. Unlike the threshold-based tag nodes, there's no confidence cutoff here - rating classification is typically a single best-guess category rather than a multi-label threshold decision, so the model just returns its top call.

Inputs and outputs

• image - the IMAGE to classify.
• model_name (default EVA02_Large) - the tagger backbone doing the rating call.

Output is a single STRING - the rating class.

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. Like the rest of the tagger family, this is a real ML model - expect a download the first time you use a given model_name, and check the pack's auto-install hook (COMFYUI_LOGICUTILS_AUTO_INSTALL=1) or requirements.txt if you hit a missing-package error.

Worth knowing

There's a companion node, GetRatingFromTextNode, that genuinely does take a different input - a filesystem path string instead of an in-graph IMAGE tensor - which makes it the one to reach for if you're scripting over a folder of files rather than working inside a live ComfyUI graph. This node, the in-graph one, is what you want for gating or tagging generations right where they're produced. Both run on WD-tagger-family models built to output Danbooru's rating vocabulary - which makes sense given this pack is aria1th's (AngelBottomless's), the same person who trains the Danbooru-trained Illustrious XL series, though this repo itself is a separate side project. Public commits stopped in January 2026, so the exact set of rating labels returned isn't documented anywhere beyond what the underlying model defines - treat it as informative, not as a hard content-moderation guarantee.