Mask_transform_sum

Most mask nodes do one thing: expand a mask, blur a mask, crop to a mask. Mask_transform_sum tries to be the one node that does all of that at once, plus turn the mask into a visible, composited image so you can actually see what you're working with instead of squinting at a grayscale thumbnail. It's a "sum" node in the pack's own naming - a bundle of the mask utility work you'd otherwise chain five separate nodes to do.

Part of ComfyUI-Apt_Preset's mask toolbox, alongside Mask_splitMask_by_color and Mask_split_mulMask. No models, no downloads - this one's pure image/mask processing, and it's easily the most parameter-dense node in that trio.

How it works

Give it a mask (and optionally a base image) and it does two jobs in sequence. First, it renders the mask as a visible style - mask_mode picks between the raw grayscale mask, a solid filled block, an outline only, a filled or outlined circle approximation, and a couple of block variants. Second, it composites that rendering onto a background chosen by bg_mode - crop to just the mask region, the full base image, a transparent background, or a flat color (white, black, red, green, blue). Between those two steps sit a pile of adjustments: grow or shrink the mask boundary before rendering (mask_expand), feather the edges (smoothness), remap the mask's intensity range (mask_min/mask_max), round the corners on block/outline renders (tapered_corners), and drop tiny stray blobs below a pixel-count floor (ignore_threshold) before any of it happens.

The other job it does, separate from styling, is cropping: turn on crop_to_mask and it'll crop the canvas down to the mask's bounding box, pad that box out with expand_width/expand_height, rescale it (rescale_crop), and snap the final dimensions to a multiple of divisible_by - which matters because most diffusion models choke or degrade on dimensions that aren't a multiple of 8 (or sometimes 16 or 64, depending on the model). That combination - crop to the masked region at a model-safe resolution - is exactly the setup you want before an "only masked" inpaint pass: full generation budget spent on the small area that actually matters, then paste it back.

Inputs and outputs

The two you'll touch first are what actually decide what comes out:

• bg_mode - background choice: crop_image (crop to the mask), image (full frame), transparent, or a flat color.
• mask_mode - how the mask itself is drawn: original (plain grayscale) is the sane default; the fill/outline/circle variants are for visualizing or stylizing rather than feeding into a sampler.

Past that, the beginner-relevant knobs are crop_to_mask + divisible_by if you're setting up a crop-and-inpaint pass, mask_expand and smoothness if your mask needs to grow a bit and lose its hard edges before it's used, and opacity for how strongly the rendered mask overlays the background image. base_image and mask are both optional inputs - you can preview a mask on its own without a background image if all you want is to check its shape.

Outputs are image and mask. Wire image into a VAE Encode if you cropped down to an inpaint-ready canvas, or just view it directly to sanity-check what your mask actually covers before committing GPU time to a run.

Installing it

ComfyUI Manager: search "ComfyUI-Apt_Preset", install, restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

Restart after. Windows has install.bat for dependencies in the repo; on Linux/Mac, pip install from the requirements file inside your ComfyUI venv. This node needs no model files - everything it does is local image processing.

Common issues

Output dimensions don't match what you expected after crop_to_mask. That's divisible_by and rescale_crop doing their job - the final size gets rounded to a multiple of divisible_by (default 8) and scaled by rescale_crop, so it won't be a pixel-exact crop of the bounding box unless both are set to leave it alone (rescale_crop at 1, divisible_by at 1). If you need an exact crop, set divisible_by to 1.

A small mask disappears entirely. Check ignore_threshold - anything below that pixel count gets dropped before rendering. It defaults to 0 so this shouldn't bite you unwittingly, but if you or a saved workflow bumped it, a genuinely small mask (a tiny blemish, a distant object) can vanish.

The rendered mask looks washed out or invisible against the background. opacity is probably low, or mask_mode is set to outline/outline_block and the outline is thin relative to the region - bump outline_thickness or switch to a fill mode if you want something you can actually see at a glance rather than a precise but faint boundary line.