[Inference.Core] Image Intensity

This one's a bit of a sleeper - it doesn't show up in most people's mental list of ControlNet conditions, but it solves a real problem: how do you recolor an image without a preprocessor that hands away the composition? The answer here is to throw away color entirely and keep only luminance. This node converts your image into a gamma-corrected intensity (brightness) map, which is what a Recolor ControlNet model conditions on - it locks in the shading and structure of the original while leaving the actual hues completely open for the prompt to decide.

Think of it as the opposite move from Canny or Lineart. Those preserve hard edges and throw away everything else; this preserves tonal/shading information (where the light and dark areas are) and throws away color and fine line detail. The result is that a recolor pass can repaint an image a completely different palette while keeping its lighting and volume intact - same shadows, same highlights, different colors.

What you actually touch: image is required. gamma_correction (default 1.0, range 0.1–2.0) is the one setting worth understanding - it reshapes the intensity curve before it's handed off. Values below 1 push the midtones brighter, values above 1 push them darker, which changes how strongly the recolor model anchors to your source's shading. If your recolored output is coming out too washed out or too contrasty relative to what you expected, this is the first thing to adjust. resolution (default 512) is the standard preprocessing size, same as every other node in this pack. The single output, IMAGE, feeds into a ControlNet Apply node - paired, as always, with an actual Recolor ControlNet checkpoint loaded separately, since this node only prepares the condition image.

One thing to be upfront about: Recolor isn't one of the original eight conditioning types from ControlNet's 2023 paper, and it isn't part of the modern union models the way canny/depth/pose are - it's a narrower, community-trained checkpoint family, so don't expect it to be sitting in whatever union model you already have loaded. You'll need to track down a Recolor-specific ControlNet checkpoint for your base model separately.

Installing it: part of LykosAI's ComfyUI-Inference-Core-Nodes pack - LykosAI is the team behind Stability Matrix, the meta-installer a lot of people use to juggle multiple local SD UIs, so the install here is more carefully packaged than a typical one-person node repo. Grab it through ComfyUI Manager (search "ComfyUI-Inference-Core-Nodes") or manually: cd ComfyUI/custom_nodes && git clone https://github.com/LykosAI/ComfyUI-Inference-Core-Nodes, then run install.py against ComfyUI's Python environment, or pip install -e .[cuda] / .[cuda12] / .[rocm] / .[directml] / .[cpu] to match your GPU backend and get an accelerated onnxruntime build, or a bare pip install -e . if you'd rather skip acceleration for now. Restart ComfyUI after.

Troubleshooting: this node is pure classical image processing (grayscale conversion plus a gamma curve), so it's not going to crash or need a model download on its own - the failure mode people actually hit is the pack failing to import in the first place, usually ModuleNotFoundError: No module named 'inference_core_nodes' because the install script ran against the wrong Python interpreter (a common trap on portable ComfyUI builds - target python_embeded explicitly). If your recolor results look flat or the model is ignoring your prompt's colors entirely, double-check you actually loaded a Recolor-family ControlNet and not a generic Canny/Depth checkpoint by accident - an intensity map fed to the wrong model type will produce confused, muddy results rather than an obvious error.