OpenPoseCanny

Canny edge maps are the workhorse of ControlNet - thin, clean outlines that tell the model exactly where the boundaries are, best for architecture, mechanical objects, and anything with crisp contours. OpenPoseCanny is the ComfyUI-CUP node that loads pre-rendered canny maps into your graph for animation work. And like everything in this pack, it doesn't run Canny detection itself; the edge detection happens in Blender, and this node just picks the right frame up.

It's an OpenPoseBase subclass with posetype = "canny". The misleading "OpenPose" prefix is just the pack's naming convention - this family is a set of frame loaders, one per control type, and this one's control type is edges. It reads <image-dir>/canny/ImageNNNN.png for the frame you request.

How it works

• image (STRING) - directory of rendered control images (filled by the Blender addon).
• frame (INT, default 0) - which frame's edge map to load.

Outputs: IMAGE (the edge map → Apply ControlNet with a canny model) and MASK (from the PNG's alpha, optional). Using renderer-provided edges has a real advantage over detecting edges from a photo: they're clean by construction. No threshold tuning, no noise floor, no wobbly lines from a video frame - just the scene's actual contours. That makes canny-from-Blender a great fit for keeping lighting and silhouette consistent across a shot.

Install and gotchas

Ships with ComfyUI-CUP. Install via Manager (search ComfyUI-CUP) or:

cd ComfyUI/custom_nodes
git clone https://github.com/AIGODLIKE/ComfyUI-CUP.git

then restart. No model downloads, no extra pip requirements.

The same three warnings that apply to all the CUP loaders: it's a loader, not a detector; it's inert in a plain web workflow (no Blender addon = no canny frames); and a missing frame silently returns a 64×64 black IMAGE/MASK that a ControlNet will turn into garbage. When edges stop behaving, check the canny folder for the actual frame file first.