Normalize Parameters

This node does nothing on its own - by design. It's a parameter generator: six float sliders on a node whose entire output is a single NORM_PARAMS value, meant to be plugged straight into the pack's Arbitrary Normalize node. If you've ever seen a custom node that wants a "normalization parameters" input and wished you could just dial it in visually, this is that dial.

Each channel (red, green, blue) gets a mean and a standard deviation, all in the 0–1 range. The mean_* values are what gets subtracted from the pixel, and the std_* values are what the result is divided by, so the math per channel is:

output = (input - mean) / std

That's exactly what Arbitrary Normalize applies. Why split the six values into their own node instead of putting sliders on the normalize node? Because you can then drive those sliders from elsewhere - a workflow that needs to try different normalization schemes without rewiring gets a single parameters wire it can swap between a few NormalizeParameters nodes, or even compute programmatically if you're building automation.

The inputs

• mean_red / mean_green / mean_blue - subtraction per channel. Defaults are all 1.0, which with default std of 1.0 means input - 1 - the "center on zero" setup.
• std_red / std_green / std_blue - division per channel. Default 1.0 (no scaling).

Each slider runs 0.0 to 1.0 in 0.1 steps. The output is parameters of type NORM_PARAMS, and it only connects to something that accepts that type - in practice, Arbitrary Normalize.

Installing and using it

It's part of the Image Misc pack. Install via ComfyUI Manager (search "Image Misc") or:

cd ComfyUI/custom_nodes
git clone https://github.com/set-soft/ComfyUI-ImageMisc
cd ComfyUI-ImageMisc && pip install -r requirements.txt

Restart, and it shows up under image/normalization. No model files, no extra dependencies beyond the pack's seconohe helper.

The honest take

If you're just trying to feed an ImageNet-pretrained model, you don't need this node - the pack has dedicated one-click nodes for the standard ranges (Normalize Image to ImageNet, [-0.5, 0.5], [-1, 1]). NormalizeParameters is for the cases in between, where the model you're feeding expects something nonstandard and you want to tune it by hand. It's the kind of small utility that makes sense to have around precisely because it's so boring. Just don't expect it to do anything until you connect the other half.