MX_Blur

MX_Blur is the boring one in the MX pack, and that's a compliment. It does one thing - a Gaussian blur - with two inputs and no mode menus, no kernel shapes, no "artistic" variants. When you just want to soften an image, blend out compression noise, or prep a background for compositing, a plain blur is exactly what you want, and this is a plain blur.

It's part of Intersection98/ComfyUI-MX-post-processing-nodes, and it sits in the same family as the pack's Kuwahara and Glow nodes - but where those are deliberately fancy, this one is deliberately basic. If you already have WAS Node Suite or Impact Pack installed you've probably got a blur too; this node's value is that it's in the same pack as everything else here, so it's zero extra installs.

How it works

It builds a real Gaussian kernel from your blur_radius and sigma and runs it through a grouped conv2d per color channel. The kernel size is 2 * radius + 1, so radius controls how wide the blur spreads while sigma controls how much the kernel falls off toward its edges. Same math Photoshop calls "Gaussian Blur," nothing exotic. One implementation detail worth knowing: sigma's relationship to radius is manual here, not auto-derived - a small sigma with a large radius produces a hollow, ring-like blur rather than a smooth one.

The inputs that matter

• image - your IMAGE tensor.
• blur_radius - 1 to 30, default 8. The spread of the blur. For subtle softening stay at 2–5; for a heavy "dreamy" look push toward the top of the range.
• sigma - 0.1 to 10, default 1.0. The kernel's falloff. Lower = a sharper-edged, more "hollow" blur; higher = a smoother, rounder one. For most uses, leave sigma around 1–3 and do your heavy lifting with radius.

Output is a single IMAGE, same resolution, unchanged channel count.

Install

Standard MX pack install. ComfyUI Manager (search MX or ComfyUI-MX-post-processing-nodes), or:

cd ComfyUI/custom_nodes
git clone https://github.com/Intersection98/ComfyUI_MX_post_processing-nodes
cd ComfyUI_MX_post_processing-nodes
pip install -r requirements.txt

Restart ComfyUI, then look under postprocessing/Filters. No model downloads - the dependencies are torch/numpy/Pillow plus opencv-python, all things your ComfyUI install almost certainly has.

Where people get burned

Two small things. First, the radius/sigma interplay: if a blur looks "weird" - sharp-edged or ringed - the culprit is usually sigma too low for the radius, not the node misbehaving. Raise sigma and the blur rounds out. Second, there's a guard in the code: a blur_radius of 0 returns the image untouched, which is a nice escape hatch rather than a bug. And since this is a bare utility with no per-node docs in the README, that's basically the whole manual. If your workflow needs an edge-preserving blur instead - smoothing flat areas without smearing edges - skip this and grab the pack's MX_KuwaharaBlur(Cartoon), which is the fancier sibling built for exactly that.