Point List to Mask 🧅

Masks are how you tell ComfyUI "change exactly this region and nothing else." Most masks are hand-painted or generated by segmentation models. This node makes masks out of points: you give it a JSON list of coordinates and it draws a circle around each one, then unions them into a single MASK. It's the second half of the pack's keypoint pipeline, and it only makes sense paired with the first half.

How it works

The point_list input expects JSON in the exact shape that the pack's OpenPoseToPointList node emits - a list of objects with x and y fields, where each value is normalized between 0.0 and 1.0. The node scales those coordinates by width and height (defaults 512×512), draws a filled circle of radius pixels around each point, and sums them all up - overlapping circles just merge, and the output is clamped to 1.0. Result: one MASK the same size as the image you intend to use it on.

So the classic setup is: OpenPoseToPointList with method set to hand → PointListToMask with a sensible radius → feed the MASK into an inpainting or masking pipeline. You get hand-shaped regions to regenerate without painting a thing. Set the method to face or all and the same chain gives you face blobs or joint blobs.

Inputs and outputs

• point_list - the JSON coordinate list (from OpenPoseToPointList, or hand-crafted - it's just [{"x": 0.5, "y": 0.5}]).
• width, height - output mask dimensions, default 512×512.
• radius - circle radius in pixels, default 50.
• Output: a single MASK.

If the point list is empty, you get an all-black (all-zero) mask rather than an error, which is lenient but means "nothing selected" - a black mask does nothing downstream, so check your pose detection if you expected blobs and got nothing.

Where you'd use it

The obvious use is hands: ask for a hand mask, feed it to an inpaint or a detail-fix step, and regenerate just the hands - a favorite trick for fixing the classic mangled-fingers problem. With the face method it's a quick face-region mask for face-focused inpainting without a face-detection model. And because the format is so simple, you can feed it coordinates from anywhere - manually typed, computed by a script, or pulled from another model - as long as they're in the same normalized JSON shape.

Installing it

It's part of ComfyUI-NegiTools - ComfyUI Manager (search "ComfyUI-NegiTools") or:

cd ComfyUI/custom_nodes
git clone https://github.com/natto-maki/ComfyUI-NegiTools
pip install -r ComfyUI-NegiTools/requirements.txt

then restart. No API key, no model downloads - but the upstream OpenPoseToPointList node does fetch pose weights on its first run, so budget for that if you're using the pair.

The honest take

Circles are a crude approximation of anatomy - a hand blob isn't a hand-shaped region, so the mask bleeds onto background and neighboring body parts. For loose regional inpainting that's fine; for surgical fixes you'll want a finer mask (segmentation or a hand-drawn one). Treat this node as the "good enough, fast, no-painting" option in the mask toolbox. It's also a genuinely nice teaching tool for understanding how normalized coordinates and masks relate - wire OpenPoseToPointList → PointListToMask → a mask preview and watch the blobs appear.