ComfyUI-CNCC (ControlNet Control Center)

English | 日本語 | 中文

ComfyUI-CNCC

Full name: ControlNet Control Center

A ComfyUI custom node that saves comfyui_controlnet_aux preprocessors as reusable "presets", so ControlNet preprocessing — and applying the result — can be handled from a single node.

Prerequisites

• comfyui_controlnet_aux must be installed.

Installation

Place this repository (comfyui-cncc/) under ComfyUI/custom_nodes/ and restart ComfyUI. No additional dependencies are required.

Features

• Run a ControlNet preprocessor on its own from a single node (outputs both the original image and the ControlNet result)
• Load a ControlNet model (pick it from control_net_name — no need to wire in a separate Load ControlNet Model node), preprocess, and apply it to positive/negative CONDITIONING, all in one node
• CNCC Settings modal for picking a preprocessor type, tuning its parameters, and saving it as a preset
  • A checkbox at save time lets you choose whether to save a thumbnail image with the preset
• Select, load, and delete saved presets from a thumbnail gallery
• Create new samples by dragging and dropping an image, including bulk-previewing it against every saved preset
• UI available in Japanese / English / Simplified Chinese (follows ComfyUI's language setting automatically)

Nodes

CNCC ControlNet (CNCCNode)

Runs a saved preset's preprocessor only. Does not apply any ControlNet.

• Inputs: cn_name (a saved preset name), image (IMAGE, optional), the node's own image-drop area (used only when image is not connected)
• Outputs: original_image, controlnet_image (the preprocessor's output)

CNCC Apply ControlNet (CNCCControlNetApplyNode)

A drop-in alternative to ComfyUI's native Apply ControlNet (ControlNetApplyAdvanced). Combines running the preprocessor and loading/applying the ControlNet model into a single node.

• Inputs: cn_name (a saved preset name), positive/negative (CONDITIONING), control_net_name (a combo of ControlNet model names from models/controlnet/), strength, start_percent, end_percent, image (IMAGE, optional), the node's own image-drop area (used only when image is not connected)
• Outputs: positive, negative (CONDITIONING), controlnet_image (a preview of the preprocessed image)
• Supports chaining multiple ControlNets in series (each new application keeps the previous one instead of overwriting it)

Troubleshooting

• DWPose preprocessor raises cv2.error: ... cv::dnn::gather: without a GPU build of onnxruntime (CUDAExecutionProvider), comfyui_controlnet_aux's DWPose falls back to running its bbox_detector ONNX model through OpenCV's DNN backend, which has a known issue with certain models (e.g. yolo_nas_*_fp16.onnx). Changing the preset's bbox_detector to yolox_l.torchscript.pt routes it through torch.jit instead, avoiding OpenCV DNN entirely and working around the bug.

License

This package (CNCC's own code) is released under the Apache License 2.0.

Licenses of dependencies and related models

CNCC does not bundle any preprocessor implementation code. It's a thin wrapper that dynamically accesses comfyui_controlnet_aux via NODE_CLASS_MAPPINGS and calls into your existing install of that package. Because of this, the license of the actual preprocessor code and weights (models) being executed needs to be checked separately:

• comfyui_controlnet_aux: Apache License 2.0. However, some of the individual preprocessors it bundles (e.g. PiDiNet, OpenPose-based ones) may carry their own license inherited from the original research implementation (research/non-commercial restrictions that differ from the package's overall license). If you're considering commercial use, check the license notice for each specific preprocessor in comfyui_controlnet_aux.
• ControlNet model files themselves (the files under models/controlnet/ selectable via control_net_name): not bundled with CNCC — they follow the license of whatever model file you've placed there (check the distributor's model card / repository).
• ComfyUI itself: GNU General Public License v3.0. CNCC only calls ComfyUI's node API (comfy.controlnet's copy()/set_cond_hint()/set_previous_controlnet(), etc.) the way any ordinary custom node does — it does not bundle or modify ComfyUI's own code.

Reference implementation

• ChrisColeTech/ComfyUI-ControlNet-Nodes (Apache-2.0): used as a reference for README structure and license notice wording.