Gaussian Blur
The blur node with a preview, because blur is rarely a surprise
- image
- ui_widget
- image
- image_list
Gaussian blur is the boring workhorse of image processing, and ComfyUI's built-in equivalents are fine. What LF_GaussianBlur adds is the pack's signature touch: a before/after compare widget right on the node, so you can slide the kernel and see it happen instead of running the graph blind. For a filter whose whole point is a visual effect, that's the feature that matters.
You reach for this everywhere - softening a mask edge, knocking down noise before a resize, defocusing a background, or pre-blurring before edge detection. It's a general-purpose tool, not a specialty node, which is exactly why having it with a live preview is more useful than it sounds.
How it works
It's a standard Gaussian blur via the pack's image-processing stack (OpenCV under the hood). Two controls do all the work:
blur_kernel_size(1–51, default 7) - the size of the kernel. Bigger kernel, more smoothing. Note it steps by 2 from 7, which keeps it odd - kernels are meant to be odd, and this node enforces that.blur_sigma(0–10, default 1) - the standard deviation of the Gaussian, i.e. how aggressive the blur is for a given kernel.
The pairing is the fiddly part. Kernel size sets the window, sigma sets the strength inside it. Small kernel + large sigma will blur harder than the kernel looks like it should, because sigma is the real intensity knob. For a soft, gentle blur on a 1024px image, try kernel 15–21 and sigma 2–3; for a heavy defocus, push kernel up and sigma toward 5+.
Outputs are image (single) and image_list (the result as a list), so it drops into both single-image and batch chains. The KUL_COMPARE widget is the highlight - before and after side by side on the node itself.
Installing it
Part of LF Nodes:
cd ComfyUI/custom_nodes
git clone https://github.com/lucafoscili/comfyui-lf
Restart ComfyUI, or search LF Nodes in ComfyUI Manager. Pure image processing, no models, no downloads beyond the pack's light Python deps (numpy, Pillow, opencv-python).
Gotchas
Don't confuse the two knobs. If blur looks weak no matter the kernel, raise sigma - kernel alone won't get you a heavy blur. And remember the step-2 constraint: the kernel widget only gives you odd values, which is deliberate (even kernels shift the result), so don't fight it by typing an even number and wondering why it snaps back.
Pack-wide note for the record: comfyui-lf is legacy, frozen since Feb 2025, with active development in lucafoscili/lf-nodes. The frozen build is stable; the compare widget got rebuilt in the new repo's LF Widgets library.
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | Input image tensor or a list of image tensors. | |
| blur_kernel_size | INT | 71–51 | Controls the size of the Gaussian blur kernel. Higher values mean more smoothing. |
| blur_sigma | FLOAT | 1.00–10 | Standard deviation for the Gaussian kernel. Controls blur intensity. |
| ui_widgetopt | KUL_COMPARE | [object Object] | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| image_list | IMAGE | — |