ImageFilterBlur

ImageFilterBlur applies a plain box blur - OpenCV's cv2.blur under the hood - to a batch of images, with separate kernel sizes for X and Y. It's the most straightforward blur in the Allor filter set: no radius guessing in Gaussian terms, no smart blur options, just "average this block of pixels and call it a day."

You feed it images (an IMAGE tensor), set size_x and size_y (both default 10, minimum 1), and get the blurred batch back, same shape, same count. That's the whole node. size_x and size_y are the kernel dimensions in pixels - bigger means blurrier, and the two axes being separate means you can blur horizontally without touching vertical detail, which is handy for streaks and motion-feel effects without actually doing motion blur.

Because it's a box blur (a mean over the kernel), it blurs fast and aggressively. It's not the soft, creamy Gaussian look people usually want for a depth-of-field vibe - for that you'd grab a Gaussian blur node. Box blur's strengths are where you want uniform, chunky smoothing: downsampling-friendly pre-processing, hiding noise or grain quickly, or softening a mask's harsh edges before it does damage downstream. And because the operation runs per-image in the batch, it works identically on a single frame or a 50-frame video strip - one of Allor's stated design goals.

The node preserves whatever channels you feed it, including a fourth alpha channel if present, which is more than you can say for some filter nodes that silently drop alpha. That makes it safe to drop into an RGBA cutout pipeline without losing your transparency.

Installing it

It ships with the Allor plugin:

cd ComfyUI/custom_nodes
git clone https://github.com/Nourepide/ComfyUI-Allor
cd ComfyUI-Allor
pip install -r requirements.txt

Restart ComfyUI, or search "Allor" in ComfyUI Manager. Two pack-level caveats: the requirements.txt drags in rembg and onnx for the segmentation nodes even though this filter needs only OpenCV (which ComfyUI already has), and Allor's repo was rebased to strip images from git history, which can break auto-updates - the docs at nourepide.github.io/ComfyUI-Allor-Doc cover how to fix that.

Real talk

ImageFilterBlur is a workhorse, not a showpiece. If you want silky bokeh, look elsewhere; if you want a fast, predictable, batch-friendly blur that won't fight you on alpha, this is it. Set both sizes to something small (3–5) for a gentle soften, or crank them for an aggressive smear. The one trap is assuming it's a Gaussian - it isn't, and using it where you need softness will leave you with visibly square-edged blur. Use it for what it is.