🕹️ CR Apply ControlNet

ControlNet is how you tell the model where things go while the prompt handles what they are - feed it a pose skeleton, a depth map, or a Canny edge map and generation follows that structure. This node applies one ControlNet to your conditioning, with a built-in switch so you can toggle it off without unplugging anything.

It's Comfyroll Studio's single-ControlNet apply node (Suzie1 / RockOfFire). If you want to stack several ControlNets, that's the Multi-ControlNet node; this one is the clean, single-condition version.

How it works

A ControlNet is a trained copy of the model's encoder that reads your control image and nudges the diffusion process toward that spatial structure. This node takes your positive conditioning, a loaded control_net, and the preprocessed control image, and folds the ControlNet's guidance into the conditioning at the strength you set. What comes out is conditioning-with-structure-attached, ready for the sampler.

One thing to be clear on: the image here is the control map - the depth map, the pose, the edge map - not your source photo. If you're doing depth or pose control, run your reference through a preprocessor first and feed that in.

The inputs that matter

• conditioning - the conditioning you're guiding (usually your positive prompt encode).
• control_net - a ControlNet model loaded with a ControlNet loader. It has to match your base architecture: an SD 1.5 ControlNet for an SD 1.5 model, an SDXL one for SDXL, and so on. No SD-era file loads on a Flux/SDXL base.
• image - the control map (already preprocessed).
• strength (default 1.0, range 0–10) - how hard the ControlNet pushes. This is the dial you'll actually tune. Community rule of thumb: ~0.3–0.7 for loose guidance, ~0.8–1.2 for strict adherence. Past that it starts overpowering the prompt; the high ceiling here is rope to hang yourself with, not a suggestion.
• switch (On / Off) - bypass the ControlNet in place. Great for A/B-ing with and without structure.

Output is the modified CONDITIONING (plus the show_help link), which goes into your sampler's positive input.

How to install it

• ComfyUI Manager: search Comfyroll Studio, install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes.git, then restart.

The pack doesn't include ControlNet models - download those separately into ComfyUI/models/controlnet.

Common issues

Ninety percent of "ControlNet isn't working" is a mismatch: the ControlNet file doesn't match your base model's architecture, or you fed it a raw photo instead of a preprocessed control map. Fix those two first. An SDXL Canny ControlNet wants an SDXL checkpoint and an actual edge map, not your original image.

If the structure is being followed too literally - stiff, traced-looking output - drop strength toward 0.5, or reach for the multi-ControlNet node if you need per-net timing. This single node applies at one strength for the whole sample; the standing community advice is often to release the condition partway through denoising so the model can add its own detail, which needs a node with start/end step controls rather than this one. And if the node's just missing, that's the Comfyroll pack failing to load, not this node.