ComfyUI-WD14-Tagger

Why we included this project

Captioning images for training data usually means exporting everything to a web tool and waiting. This ComfyUI extension keeps the whole job local: it runs SmilingWolf's WD 1.4 tagger models inside the node graph, spitting out booru-style tags from any image. Models download and cache themselves on first use, and you can swap between backbones like ConvNextV2 or MOAT, with separate thresholds for regular and character tags so you control how much noise gets through. It also works as a right-click shortcut on any image node, which makes batch captioning a matter of a few clicks rather than wiring up extra nodes. Once the weights are cached it runs fully offline, a real advantage when you're processing large or sensitive datasets without touching a cloud API.