Apply ControlNet (full-pipe)

ControlNet is how you tell the model where things go - pose skeletons, edge maps, depth - instead of just what they are. In a stock ComfyUI graph that's a handful of nodes bolted onto the conditioning path. Apply ControlNet (full-pipe) folds the whole thing into this pack's pipe convention: full_pipe in, pick a ControlNet model and an image, and the positive conditioning inside the pipe comes out carrying the control hint. If you skip the image or the model, it's a clean passthrough - the pipe comes out unchanged.

Mechanically it does what ComfyUI's ControlNetApplyAdvanced does under the hood: loads the .safetensors from your models/controlnet/ folder, attaches it to the pipe's positive conditioning with a strength and a start/end window, and marks it to also apply to the uncond pass. One nicety the source keeps: if the conditioning already carries a previous ControlNet, the new one chains onto it rather than replacing it, so you can stack control nets by running this node twice.

The inputs that matter

• control_net_name - dropdown of installed ControlNet models. Architecture matters: a ControlNet trained for SDXL won't run on a Qwen/Anima base, and the pack's own docs note that post-SDXL control support is patchy. For Anima specifically, the community's ControlNet story is thin - the LLLite ports are weak on the 1.0 base, and a control LoRA often works better.
• image - the hint image. Preprocess it first (canny, depth, openpose) with whatever you like; this node just feeds the raw image in.
• strength (default 1.0) - how hard the condition pushes. The KB's standing advice: 0.3–0.7 for loose guidance, 0.8–1.2 for strict.
• start_percent / end_percent (0 to 1) - when during denoising the control applies. The classic move is end_percent around 0.5–0.8: let the control set the composition, then let the model add its own detail in the later steps.

Outputs are the updated full_pipe and the image you fed in, passed back out for convenience.

Installing it

It's in mudknight-utils - ComfyUI Manager (search comfyui-mudknight-utils) or:

cd ComfyUI/custom_nodes
git clone https://github.com/mudknight/comfyui-mudknight-utils

No Impact Pack required. You need the ControlNet model file itself in ComfyUI/models/controlnet/.

Where people get burned

The passthrough behavior trips people up: if you forget to wire an image, the node silently does nothing rather than erroring, so your "controlled" render isn't controlled at all. Check the image output if the pose isn't showing up. And the ControlNet-to-model mismatch is the other classic - a control file that loads but has no visible effect is almost always the wrong architecture for your base model. Match families, keep the end_percent under 1.0 for most jobs, and it'll do what it says.