Shima Gaussian Blur
The plain, dependable blur — for softening, glow, and depth-of-field fakes
- image
- IMAGE
Sometimes you just need to blur an image, and ComfyUI's core doesn't hand you one. Shima.GaussianBlur is that node: it applies a Gaussian blur with one strength slider and an on/off switch. Nothing fancy, and that's the point - it's a utility you drop into a post-processing chain when you need a soft pass, and it gets out of the way.
Before you use it, know what Gaussian blur actually is, because it's the most misused blur. A Gaussian kernel smears everything - edges included - which makes it the tool for softening an image, building a bloom/glow pass (blur a bright area, then screen it back over the original), faking shallow depth of field on a background, or generating the reference copy for an unsharp mask. What it is not is a skin-smoother: that job wants an edge-preserving bilateral filter, and reaching for Gaussian there turns your subject to mush (the post-processing doc walks through exactly this distinction). Use it for what it is and it's great.
The mechanism is plain Pillow: the image goes through ImageFilter.GaussianBlur on the CPU with radius=strength. That tells you two things. One, it's a real, standard Gaussian - predictable results. Two, it's CPU and per-image, so it's fine for a single pass but not what you want if you're writing a loop over a thousand frames (there are GPU blur nodes for that).
Inputs:
- image (IMAGE) - what to blur.
- strength (FLOAT, 0–200, default 1) - the blur radius in pixels. 1–3 is a gentle soft-focus; 10+ is dreamy mush; 50+ is an abstract smear, which is actually what you want for a background-bokeh plate.
- enabled (BOOL, default on) - the kill switch. Off, and the image passes through untouched. Wire a toggle to it and you can A/B a blurred and sharp path without rewiring.
Output: a single IMAGE.
The one honest caveat: strength is a radius, and on a 1024px image a radius of 1 is barely perceptible. Beginners set 1, see "nothing happened," and conclude it's broken. If you want to see it, start around 5–8 and adjust from there.
Install the pack via ComfyUI Manager (search "Shima") or
cd ComfyUI/custom_nodes
git clone https://github.com/KDB-USJP/shima_wf
then restart. No models, no extra deps beyond Pillow, which ComfyUI already has.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| strength | FLOAT | 1.000–200 | — |
| enabled | BOOLEAN | true | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |