Mask Rectangular Area 🦾

Mask Rectangular Area draws a rectangle. That sounds too simple to need a node, until you need a rectangle mask - a region where a model is allowed to generate (or forbidden from touching) - and you need it to be exact, reproducible, and numerically aligned to your canvas. That's what this is: a mask tensor with a solid white rectangle at coordinates you type in.

This is the same idea as ComfyUI's built-in rectangle mask tools, but standalone and pixel-addressable: you specify x, y, width, height for the rectangle and image_width/image_height for the canvas it lives on, and out comes a mask. If you've ever set up regional prompting or a "keep this region untouched" inpaint workflow with a fixed rectangle - a text box, a face area, a logo zone - you know why you want it.

How it works

Under the hood it creates a zeroed tensor of image_height × image_width, then sets the slice from (x, y) to (x+width, y+height) to 1. The result is a single-channel MASK. Slicing past the edge is safe - torch just clips it - so a rectangle that hangs off the canvas gets cut at the boundary rather than erroring.

The step on every input is 64, which is deliberate: it keeps rectangles aligned to the multiples ComfyUI's latent sizes and many models prefer. If you need sub-64 precision, you can still type in an exact number; the step only affects dragging.

The inputs that matter

• x, y (required, default 0) - top-left corner of the rectangle.
• width, height (required, default 512) - rectangle size.
• image_width, image_height (required, default 512) - the canvas dimensions the mask must match downstream. This is the one beginners get wrong: the mask is only useful if these match the image/latent you're conditioning.

Output: MASKS (a MASK tensor).

Install

Part of ComfyUI-Beyond_nodes. ComfyUI Manager → search "Beyond_nodes" → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/beyondprompting/ComfyUI-Beyond_nodes

Restart. No extra dependencies.

Common issues

• The mask is the wrong size / doesn't line up. image_width/image_height define the mask's dimensions - they must equal your actual image dimensions, not the rectangle's. A 512×512 mask on a 1024×1024 image is useless.
• It's a hard edge and I want a soft one. This node doesn't blur. Pipe the mask through a Blur Mask node, or pair it with the pack's MaskBoundingBox, which adds padding and blur for exactly this.
• I want a mask from an existing image's subject, not a rectangle. Different tool - use a segmentation/SAM workflow. This node is for when you know the geometry up front.

Rectangle masks are the boring workhorse of controlled generation: fixed regions, reproducible results, no model involved. When the region you care about is a rectangle, this is the fastest way to build it.