Mix Images By Mask

Mix Images By Mask does exactly one thing: blend two images together using a mask as the interpolation weight. Where the mask is white you get image2, where it's black you get image1, and anywhere in between you get a proportional mix. It's the general-purpose sibling of Mix Color By Mask - instead of blending a flat color into an image, you're blending a whole second image in.

This is the node that quietly does the compositing in a lot of masked-editing workflows, and it's often a cleaner alternative to Paste By Mask when you don't actually need the crop/resize logic. If you've already got two full-size images - say, a generated version and the original - and you just want to combine them along a mask without worrying about region sizing, this is the more direct route. It's also the fix people reach for when Paste By Mask's resize behaviors get finicky: mix the full images directly instead of cutting and pasting a sub-region.

The inputs that matter

• image1 - the base image, shown where the mask is black (or low).
• image2 - the image blended in, shown where the mask is white (or high).
• mask - the interpolation weight, per-pixel.

The single output is the blended image.

Because the blend is proportional to the mask's actual value, a soft, feathered mask gives you a smooth crossfade instead of a hard cut - no separate feathering step needed if your mask already has soft edges from a Gaussian Blur or from ClipSeg's natural output. If you want a hard edge instead, round the mask first with Combine Masks (round_result on) or Mask Morphology.

Installing it

Mix Images By Mask is bundled with Masquerade Nodes. Install through ComfyUI Manager (search "Masquerade Nodes") or manually: cd ComfyUI/custom_nodes && git clone https://github.com/BadCafeCode/masquerade-nodes-comfyui, then restart ComfyUI. It's pure pixel math - no models to download, no extra Python dependencies to install.

Common issues

The one real requirement is that image1, image2, and mask all need to be the same resolution - this node blends pixel-for-pixel and doesn't resample anything for you. If you're combining a generated crop with a full-size original, resize the crop back up first (or work with Cut By Mask / Mask To Region / Paste By Mask instead, which handle sizing explicitly). If the blend looks reversed - image2 showing up where you expected image1 - check whether your mask is inverted; a common cause is feeding in a mask meant for "protect this area" when the node expects "reveal this area," which are opposite conventions and easy to mix up when chaining several mask nodes together.