Every ControlNet preprocessor, ported to ComfyUI in one pack

What it does This repo wraps the annotator code from lllyasviel’s original ControlNet repo as ComfyUI custom nodes, so you can generate ControlNet hint images — canny edges, depth maps, OpenPose stickmen, scribbles, segmentation masks — directly inside your ComfyUI workflow. Weights are pulled from the 🤗 Hub. It covers line extractors (Canny, HED, lineart, TEED, AnyLine), depth and normal estimators (MiDaS, Zoe, Depth Anything V2, Metric3D, DSINE), pose estimators including an animal variant, segmentation, recolor, and even an optical flow node for DragNUWA.

The interesting bit The README is refreshingly blunt: the code is “copy-pasted” from the original annotator folders, with all credit to lllyasviel. It’s glue code, but glue that saves you from reimplementing a dozen model integrations — and the node tables map each preprocessor to the specific ControlNet/T2I-Adapter checkpoints it pairs with, which is genuinely useful reference material. There’s also an AIO Aux Preprocessor node that bundles everything except Inpaint into one dropdown, at the cost of not exposing per-preprocessor thresholds.

Key highlights

• Dozens of preprocessors spanning lines, depth, normals, poses, faces, segmentation, recolor, and optical flow
• AIO Aux Preprocessor node for quick switching without rewiring the graph
• OpenPose-format JSON is exposed via app.nodeOutputs and the /history API endpoint, so pose editors and API consumers can grab keypoints per frame
• Compatibility tables cross-reference each node against sd-webui-controlnet equivalents and matching ControlNet checkpoints
• Includes newer additions like Depth Anything V2, TEED soft edges, and Metric3D

Caveats

• The AIO Aux Preprocessor convenience node can’t set a preprocessor’s own threshold parameters — you must drop in the dedicated node for that
• It’s a wrapper, not original research; upstream bugs in the annotator code presumably ship along with it
• Animal pose JSON uses a “kinda similar” (their words) format to OpenPose JSON, not the standard one

Verdict If you run ControlNets in ComfyUI, this is close to mandatory infrastructure. If you don’t use ComfyUI or ControlNet, there’s nothing here for you.