CogVideoControlNet

This is the node that actually uses a CogVideoX ControlNet. The loader node ((Down)load CogVideo ControlNet) gets the ControlNet weights into memory; this node combines those weights with your control images and produces the conditioning bundle that the sampler consumes. Think of it as the "apply" step to the loader's "load" step - you need both.

The idea is the standard ControlNet bargain applied to video: you hand the model a per-frame structural guide - a sequence of canny edge maps or HED sketches - and it generates content that follows that structure while the prompt fills in style and detail. That's how you get motion locked to a specific shape or silhouette instead of whatever the model would freely invent.

The inputs that matter

• controlnet - the COGVIDECONTROLNETMODEL from the loader node. This is the required upstream connection.
• images - your control image sequence. These are the edge maps (canny) or soft-edge maps (HED) the model follows, one per frame. You typically make them with a preprocessor pack that converts a source video into edge sequences. The control type has to match the ControlNet you loaded - canny maps for a canny ControlNet, HED maps for a HED one.
• control_strength (default 1) - how hard the ControlNet enforces the structure. 1 follows the control tightly; lower it if the guide is fighting the prompt or making motion look pasted-on. This is your main tuning knob.
• control_start_percent (0) / control_end_percent (1) - the window of the denoising schedule the control is active over. The defaults apply it the whole way. A common trick: end the control early (e.g. 0.5) so structure is set in the first half and the model refines freely after, which loosens the rigidity.

The output, cogvideo_controlnet (a COGVIDECONTROLNET), plugs into the controlnet slot on CogVideo Sampler.

How to install it

Part of the wrapper:

• ComfyUI Manager - search ComfyUI-CogVideoXWrapper, install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-CogVideoXWrapper, then pip install -r ComfyUI-CogVideoXWrapper/requirements.txt, restart.

For generating the canny/HED control images, grab a preprocessor pack like comfyui_controlnet_aux.

Common issues & troubleshooting

The output ignores the control. Two usual causes: the cogvideo_controlnet output isn't wired into the sampler's controlnet slot, or control_strength is too low. Also make sure your images are actual edge/sketch maps, not the raw source frames - feeding plain photos where the model expects a canny map won't guide anything useful.

The result looks stiff or traced. control_strength too high, or the control runs the whole schedule. Lower the strength, or pull control_end_percent back so the model gets free rein in the later steps to add natural detail and motion.

Wrong control type. A HED ControlNet expects HED maps; a canny ControlNet expects canny edges. Mixing them - canny maps into a HED model - gives muddy, confused guidance. Match the preprocessor to the ControlNet you loaded.