Superside Mask Preview

A MASK is a tensor of numbers between 0 and 1, and ComfyUI's node graph does not show you tensors - it shows you images. So every time a mask comes out of a detector or a selector and you have no idea what it actually covers, you need a preview node. This is the Superside version of that: Superside Mask Preview converts a mask to a grayscale image and shows it right in the graph, with no extra wiring and no save step.

It's the in-house equivalent of comfyui_essentials' MaskPreview+, and it does the one job faithfully. White in the preview means mask value 1.0 (fully selected), black means 0.0 (untouched), gray means partial. That gray band is more useful than people give it credit for - partial-opacity masks are how Superside's own portrait pipeline does soft exclusion, and being able to see the soft edge is what tells you whether it's right.

The inputs and outputs

One input: mask. That's it. You feed it any MASK tensor - from a mask editor, a region selector, a segmentation node, anywhere.

The output is images, a grayscale IMAGE version of the mask. The node is flagged as an output node, meaning it's meant to be a terminal preview in the graph - but the image output is a real tensor, so you can also grab it and use it elsewhere (e.g. to save a PNG of a mask for debugging or documentation).

How it works

The conversion is the same routine as this pack's Mask To Image node: flatten the batch, move the channel into the right position, and broadcast the mask value to three channels (R=G=B=gray). Then it saves the result to ComfyUI's temp folder and shows it as a UI preview. Pure local math, no API, no key, no model.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes
pip install -r requirements.txt

Restart ComfyUI, and it's under Superside. No configuration beyond that.

Why you'll actually use this

Masks are the step where workflows silently go wrong, and they're invisible while they do it. A region selector returns a mask that's flipped, tiny, or covering the wrong object - and you won't know until the final composite looks wrong. Dropping a preview node on every mask in your graph turns "debug at the end" into "spot it immediately." It's the cheapest debugging tool in the pack, and the one people skip until the first time a mask betrays them.