Mask Combiner (RMBG) 🎭

You've got two masks and you need one. Maybe one from a background remover and one from a YOLO face detector, and you want them merged. Or you want the overlap of two, or one with a chunk cut out of the other. Mask Combiner does the boolean algebra: union, intersection, difference, on up to four masks at once.

This is plumbing, and good plumbing is invisible. In any real segmentation workflow you end up with masks from different sources - a subject mask here, a hair-detail mask there, a region you manually painted - and combining them cleanly is the difference between one tidy node and a spaghetti of image-math nodes. Mask Combiner is that one tidy node.

How it works

It takes your masks and applies a set operation to them, pixel by pixel. Three operations, and they mean exactly what they say:

• combine - union. Everything masked in any input ends up masked. This is the default and the one you'll use most: stack a body mask and a hair mask, get the whole subject.
• intersection - only the pixels masked in all inputs survive. Use it to constrain - "the part of the subject that's also inside this region."
• difference - subtract later masks from the first. Mask A minus mask B. Great for punching a hole: take a full-subject mask, subtract the face, and inpaint everything but the face.

The inputs and outputs that matter

• mask_1 (required) - your base mask.
• mask_2, mask_3, mask_4 (optional) - up to three more to combine with it. Connect as many as you need; unconnected ones are ignored.
• mode - combine / intersection / difference, as above.

One output: MASK. Wire it wherever a mask goes next - an extractor, an inpaint conditioning node, another combiner if you're doing something baroque.

How to install it

• ComfyUI Manager: search Comfyui-RMBG, install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/1038lab/ComfyUI-RMBG, then pip install -r requirements.txt, and restart.

Pure math, no model - instant.

One thing to watch

difference is order-sensitive: it's mask_1 minus the rest, not a symmetric operation. If you're getting the inverse of what you expected, you've probably got your masks plugged into the wrong slots - swap mask_1 and mask_2. The other gotcha is size: masks of different dimensions don't combine meaningfully, so make sure everything upstream is working at the same resolution before it reaches here. If your masks come from different-sized sources, resize them to match first.