Blur images

Blurring is a surprisingly common pipeline step: softening a background layer, defocusing a reference before it gets fed to a detailer, pre-blurring for a stylized look, or anonymizing a batch of faces before upload. LF_BlurImages is the LF Nodes take - and its twist is that it's built for batches of images, and it tracks filenames while it works.

The core control is blur_percentage, from 0 to 1. That's the whole clever bit: instead of a pixel radius you have to tune per resolution, the blur scales with the image's own dimensions. 0% is no blur, 100% is a maximum blur "based on the image's dimensions" that renders a 1024px image basically unrecognizable - a single soft blob. Because it's relative, the same blur_percentage behaves consistently across a mixed batch of image sizes, which a fixed-radius blur never does.

The inputs and outputs that matter

• image - a batch of images (it says "list of images," but it also accepts a single tensor).
• blur_percentage - 0 to 1, default 0.25. Small values for a gentle soft-focus, big values for heavy defocus.
• file_name (optional) - a list of names corresponding to the images. Give it the names from LF_LoadImages and you get them back on the file_name output, in order, after blurring.
• Outputs: image, image_list, file_name (list), and count (how many images came through). count is handy for sanity-checking that your whole batch survived the trip.

That filename pass-through is the quietly useful bit. If you're processing a dataset - blur the reference set, keep the names paired with the outputs - you can wire the blurred result and its names onward together and never lose track of which blob was which image.

How it fits the LF pack

It's the batch-oriented sibling of the pack's other image filters (brightness, contrast, clarity). The masonry gallery widget (KUL_MASONRY) shows all your blurred results on the node body at once, which is much better than previewing blurred images one at a time. It's a straightforward torch + PIL blur under the hood, so it's fast even on big batches.

Installing it

Same pack, same drill:

cd ComfyUI/custom_nodes
git clone https://github.com/lucafoscili/comfyui-lf

or ComfyUI Manager → LF Nodes → install → restart. No dependencies, no model downloads. It's the frozen legacy repo (development moved to lf-nodes), so expect stability over new features.

Where people get burned

• blur_percentage is relative, so "0.1" is not "a little bit" - it's 10% of a dimension-scaled maximum. Start at 0.05–0.15 and work up; going straight to 0.5 on a big image is a lot more blur than it looks like on a small preview.
• If you feed it a single image and expected one output, remember image is still a batch of one - fine for most downstream nodes, but a list-consuming node downstream may want image_list.
• file_name is optional, and if you don't supply it, you get back an empty list rather than generated names. If you care about names, wire them in from LF_LoadImages.

For dataset work and any batch-softening job, this is the node you'll actually reach for in this pack.