OpenCV GaussianBlur_1

You already know how this goes: this pack auto-generates one node per OpenCV overload, and GaussianBlur_1 is the duplicate of GaussianBlur_0. Identical code, identical inputs, identical output - the two stubs in the current release do the same thing. So use whichever. The only thing that differs is the badge on the node.

Which leaves one genuinely useful job for this page: making sure you know how to drive the node, because it's one of the few in this pack you might reach for every day.

What it does and what to set

It's cv2.GaussianBlur - a proper Gaussian blur that smears everything, edges included. The classic uses: softening a frame, building a bloom/glow pass by blurring a bright copy, or generating the blurred reference for an unsharp mask. It's the deterministic pixel-op layer, not a model in sight.

The inputs:

• src (NPARRAY) - the image, via Image2Nparray.
• ksize (STRING) - kernel size as a Python literal, [5, 5]. Must be odd and positive; even values throw an OpenCV assertion.
• sigmaX / sigmaY (FLOAT) - the blur's standard deviation. sigmaX = 0 makes OpenCV derive it from ksize, which is usually what you want. sigmaY = 0 mirrors sigmaX.
• borderType (INT) - edge handling; 0 (default) is right for images.
• hint (INT) - an OpenCL hint from OpenCV 4.11; leave 0.
• dst (NPARRAY, optional) - ignored out-parameter, per the README.

Output is nparray, the blurred image.

Installing

Ships in geroldmeisinger/opencv-comfyui. ComfyUI Manager: search "opencv-comfyui". Or:

cd ComfyUI/custom_nodes
git clone https://github.com/geroldmeisinger/opencv-comfyui

Restart ComfyUI. The pack's requirements.txt installs opencv-contrib-python, numpy, and torch.

Common issues

Most invalid syntax errors on this node are the ksize literal - it must be a valid Python tuple/list, [5, 5] or (5, 5), not 5,5. Even-sized kernels throw the must be odd assertion. And the pack-wide rules stand: single-image batches only, BGR in and out (Comfy is RGB, this is OpenCV), grayscale images where a color image is expected trigger the CV_8UC1 assertion - convert first with cvtColor (code=6 for BGR2GRAY).

Final thought, and it's the recurring theme of this pack: the _0/_1 duplication is a feature of the generator, not something you did wrong. The author's own README calls these nodes ugly and asks for feedback. If you only ever use one Gaussian blur node, this is a fine one to keep - it's the real OpenCV implementation, and it's free.