Nodes/radiance/◎ Radiance Control
ComfyUI Node

◎ Radiance Control

A ControlNet apply node with a Union mode picker built in

By fxtdstudios·Created 7 months ago·Updated 9 days ago· 242
◎ Radiance Control
  • conditioning
  • control_net
  • image
  • conditioning
strength1.00
start_percent0.00
end_percent1.00
control_typeauto

The stock ComfyUI ControlNet apply node works fine for the classic one-condition-per-model case. Radiance Control is what you reach for when your ControlNet is a Union model - one of those combined checkpoints (Flux's union, xinsir's SDXL union, the Alibaba PAI unions) that bundle canny, depth, pose, and half a dozen other modes into a single file and need to be told which mode to activate. Radiance Control has a dropdown for exactly that, which saves you from reaching for set_extra_arg hacks or a second custom pack.

At its core it's the same mechanism you already know from any apply node: it takes your conditioning, a control_net, and the conditioning image, and injects the ControlNet as a control entry on the conditioning so the sampler steers spatial structure - edges, depth, pose - during generation while the prompt decides what the content is.

The inputs that matter

  • conditioning - your positive conditioning from a text encode.
  • control_net - the ControlNet model. Leave it unconnected or set strength to 0 and the node bypasses gracefully, passing your conditioning through untouched instead of crashing. That's a small thing that saves real pain when you're toggling a condition on and off.
  • image - the preprocessed condition image (canny edge map, depth map, pose skeleton - preprocess it first, the node doesn't do that for you).
  • strength - 0 to 10, default 1. How hard the condition steers. Backing it off to 0.6–0.8 is the classic fix for "ControlNet is fighting my prompt."
  • start_percent / end_percent - when control kicks in and cuts out, 0.0–1.0. Start later or end earlier to let the model do its own thing for part of the denoise.
  • control_type - the Union mode selector: auto, openpose, depth, hed/pidi/scribble/ted, canny/lineart/anime_lineart/mlsd, normal, segment, tile, repaint. On a single-purpose ControlNet, "auto" is right and the rest is ignored.

Output is a single conditioning, wired straight into your sampler. The node also chains properly for multi-ControlNet stacks - it respects a previous control on the conditioning, so you can daisy-chain several of these.

Honest positioning

You don't strictly need this node - the core Apply ControlNet plus a bit of fiddling covers most single-ControlNet work. Where it earns its place: the Union dropdown, the graceful bypass, and the start/end scheduling without a scheduler node. It's a quality-of-life wrapper, and it does its one job cleanly. The pack's launch thread drew real skepticism about the broader "32-bit pipeline" claims, but this node doesn't make that claim - it's just a tidy ControlNet applicator that happens to live in a fancier pack.

Install

It's in the fxtdstudios/radiance pack - ComfyUI Manager search "Radiance", or:

cd ComfyUI/custom_nodes
git clone https://github.com/fxtdstudios/radiance.git
cd radiance
pip install -r requirements_linux.txt   # or windows/mac_silicon variants

Restart after installing. Note this node needs a ControlNet model file - the pack does not download one, so you still need to drop your own .safetensors in models/controlnet. The pack's dependency install (OpenEXR, OpenColorIO, colour-science, transformers) is heavy and makes the first boot slow; if Manager's security level refuses the install, clone manually per the README.

CategoryFXTD Studios/Radiance/Generate

Inputs (7)

NameTypeDefaultDescription
conditioningCONDITIONING
control_netCONTROL_NET
imageIMAGE
strengthFLOAT1.000–10Global strength of the control effect.
start_percentFLOAT0.000–1Percentage of the generation where control starts (0.0 = beginning).
end_percentFLOAT1.000–1Percentage of the generation where control ends (1.0 = end).
control_typeCOMBOautoFor Union ControlNets (like Flux), select the specific control mode (Canny, Depth, etc.).

Outputs (1)

NameTypeDescription
conditioningCONDITIONING