LayerFilter: GaussianBlur

This is the node you reach for when you just want to blur an image and don't want to think about it. No radius-in-weird-units, no kernel size, no separate horizontal/vertical passes. Image in, one number, blurred image out. That's the whole pitch, and honestly for 90% of blur jobs that's all you need.

Gaussian blur is the classic soft, even blur - it weights nearby pixels on a bell curve, so you get a smooth falloff instead of the boxy look you get from a cheap average. In a ComfyUI graph you'd use it to fake depth of field on a background, soften a plate before compositing something on top, knock the sharpness off a mask-derived layer, or just take the crunch out of an over-sharpened generation.

How it works

It runs a standard Gaussian convolution over the whole frame. The blur value is the radius - bigger number, wider the blur spreads, mushier the result. It's a global filter: it hits every pixel equally and doesn't know or care about any mask. If you want to blur only part of the image, blur the whole thing and composite it back through a mask (LayerStyle has plenty of nodes for that), or blur a masked region upstream.

The inputs and outputs that matter

There are only two inputs, which is the point:

• image (IMAGE) - what you're blurring.
• blur (INT, default 20, range 1–999) - the radius. 20 is a gentle softening. Push it to 60–100+ for that fully-out-of-focus background look. Below ~5 you'll barely see it.

Output is a single image (IMAGE) you wire straight into whatever's next - a Save Image, a composite, another filter. Nothing surprising.

How to install it

Easiest path is ComfyUI Manager: hit the Manager button, search ComfyUI Layer Style (or just LayerStyle), install, restart ComfyUI. Manual route if you prefer the terminal:

cd ComfyUI/custom_nodes
git clone https://github.com/chflame163/ComfyUI_LayerStyle

Then restart so the pack's requirements.txt gets picked up. GaussianBlur itself is pure image processing - no model downloads, no heavy weights, it's basically a wrapper around a standard blur.

Common issues & troubleshooting

The classic snag with this whole pack isn't this node, it's the pack refusing to load at all - you'll see IMPORT FAILED for ComfyUI_LayerStyle in the startup log and none of its nodes appear in the menu. Two usual causes. One: you've got an old install from back when this was a single giant pack. It got split into ComfyUI_LayerStyle and ComfyUI_LayerStyle_Advanced to cut down the dependency weight, and a stale combined copy can choke on load - delete the folder and reinstall clean (this fix has a long trail of "it worked, thank you" replies behind it). Two: a broken dependency further up your Python environment, classically a mangled transformers/tensorflow combo, throws during import and takes the pack down with it. Manager's "Try Fix" sometimes clears it; a clean delete-and-reinstall almost always does.

For the node itself there's not much to break. If the output looks identical to the input, your blur is too low - bump it. If a big blur is slow on a huge image, that's just the radius; Gaussian cost scales with it. And remember it's whole-image: if you expected only the subject or only the background to blur, that's not this node's job - you need to mask and composite.