Skip Head Mask v5.1.0

The name is a lie, in a good way. Skip Head Mask doesn't skip a head mask - it cuts the head out of your image, full stop. Feed it a photo and a head mask, and it blackens everything above the bottom of the mask, leaving you with a torso-only version of the shot. One node, one output, done.

This is a virtual try-on utility from the tri3d-comfyui-nodes pack (TRI3D-LC, v5.1.0). In a try-on graph you often don't want the inpaint pass touching the face at all - the head should stay exactly as it is while the torso gets regenerated. The classic way to enforce that is to hand the sampler a blacked-out head region plus a matching mask, and this node is the shortcut for producing that blacked-out image from a segmentation mask you already have.

How it works

Plain OpenCV, no model involved. It takes image and head_mask, finds the lowest point of the head mask's contour (the chin, roughly), and zeroes out every row above it:

• image → the photo you're cutting the head out of.
• head_mask → a white-on-black mask covering the head. Any segmentation source works - the pack's own ATR/extract-masks nodes, BiRefNet, whatever you already use.

Output is a single IMAGE: the original with the head region blacked to zero. That's it. There's no feathering, no blur, no alpha - it's a hard cut you're expected to feed into the rest of your inpainting chain.

Where you'd actually use it

The common move is to pair it with a matching mask and run the blacked image through inpainting (or feed it as the "don't touch" reference in a two-pass try-on). Because it hard-cuts to black rather than writing a soft mask, wire it into nodes that understand blacked-out regions - or convert to a mask yourself if your inpaint setup expects MASK input.

Install

ComfyUI Manager → search tri3d-comfyui-nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/TRI3D-LC/tri3d-comfyui-nodes

Restart ComfyUI. Nothing to download - this node needs only cv2/numpy, which ship with ComfyUI.

Gotchas

• Batch behavior: the node processes only the first frame of a batch (image[0]), so don't feed it a video or a multi-image batch expecting per-frame output.
• Hard cut: if your head mask is slightly small, you'll keep a sliver of face; if it's too big, you lose neck. Check the chin boundary before you rely on the result.
• Empty mask: no contour means no blackout - you get the image back unchanged, which is a silent failure if your upstream segmentation died.

It's a boring, honest utility node from a pack with a near-empty README, and that's fine. When you need a headless torso, it's faster than building the crop-and-blackout yourself.