gaussian blur

The Gaussian Blur_Ally node (displayed as "gaussian blur") does exactly what it says: blurs an image with a gaussian filter and hands it back. It's one of three nodes in this pack ported from "theAlly" - a well-regarded single-purpose node author whose own nodes were apparently fiddly to install, which is why Aegis72 folded them in here with an enabled switch bolted on. The extension matters more than it sounds: you can leave the node permanently in your workflow and flip it off instead of deleting and re-adding it every time you A/B a sharp versus soft look.

How it works

The mechanism is plain PIL: the image becomes a PIL image, gets ImageFilter.GaussianBlur applied with the strength value as the radius, and converts back to a tensor. Worth knowing the implementation detail the original author themselves noted - it's technically a gaussian approximation (PIL's GaussianBlur is built on repeated box blurs), but for anything short of pixel-peeping, it looks and behaves like a real gaussian.

Inputs and outputs

• image (IMAGE, required) - the image to blur.
• strength (FLOAT, default 1.0, range 0–200) - the blur radius. Default 1.0 is a gentle softening; you'll want single digits for most effects and double digits only for heavy backgrounds.
• enabled (BOOLEAN, default true) - flip to false and the image passes through untouched.
• Output: IMAGE.

When you'd use it

Pre-save finishing, mostly - softening a harsh edge, dropping detail before an upscale or output, or building a depth-of-field effect by blurring a background layer you composite under a sharp foreground. It handles a single image or a batch, since it just applies the filter to each frame.

Installing it

Standard for the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/aegis72/aegisflow_utility_nodes

Restart ComfyUI, or install "AegisFlow Utility Nodes" via ComfyUI Manager. No model downloads; this is PIL over the image tensor, nothing heavy.

Where people get burned

The main surprise is scale: because strength is the PIL radius in pixels, values that feel small on a 1024px image behave differently on a 512px one - a radius that looks subtle upscaled can be a smear at lower resolution. And keep in mind this is a blunt tool: it blurs everything, no masking, no edge-aware magic. If you need selective blur you'll want a proper depth-based or masked approach; this node is the simple version that just works.