HT Mask Dilate

The name says "dilate," but the thing this node actually does is more useful: it finds where the non-zero pixels in your mask are, crops the image (and mask) to that bounding box, and resizes the crop to a standard dimension - reporting the scale factor it used. If you've got a subject isolated by a mask and you want to process it at a consistent size without the dead space around it, this is the node.

How it works

Three steps, all grounded in the source:

1. Find the bounds. It scans the mask for non-zero pixels and computes the tight bounding box around them (with padding, default 64, breathing room around the content).
2. Crop. Both the image and the mask get cut to that box, aligned.
3. Scale to a bucket. The cropped long edge is resized to a "standard" size chosen by scale_mode:
   • Scale Closest (default) - nearest standard bucket to the crop size; least distortion, keeps things roughly as big as they were.
   • Scale Up - only ever scales up to the next bucket (good for making a small subject big enough to detail properly).
   • Scale Down - only scales down (for fast previews or model input constraints).
   • Scale Max - scale up to the largest bucket.

Outputs: dilated_mask and cropped_image (both BHWC, ready to feed the next node), plus width, height (the final crop dimensions), and scale_factor - the float you can multiply a coordinates/width by if anything downstream needs to translate positions back to full-frame space.

What it's for

The classic use: masked subject → this node → detail/upscale pass → paste back. You detect a person, crop them out at a clean standard size, run a high-quality pass on just them, and use the scale factor to place them back correctly. It's also the "extract the subject from the background" step for layer workflows - a natural feed into the pack's Layer Collector, where the mask becomes the alpha. And because it normalizes size, it's how you make differently-sized masked subjects uniform before a batch pass.

Two inputs matter and the rest is taste: mask (the content finder) and scale_mode (what "standard size" means to you). The KB's inpainting essay covers why masking + cropping is the pattern behind most local edits - this node is that pattern's plumbing.

Installing

Standard pack install - Manager → "HommageTools for ComfyUI", or:

cd ComfyUI/custom_nodes
git clone https://github.com/ArtHommage/HommageTools.git
cd HommageTools && pip install -r requirements.txt

restart. No models, no extra dependencies.

The honest take

Solid, well-scoped utility - but mind two things. First, it handles one contiguous region per run; if your mask contains multiple separated blobs, the bounding box will span them all and include the gaps, and you'd want the pack's Multi Mask Dilate variant for region-per-blob processing. Second, it's easy to conflate "dilate" with mask expansion - this node doesn't grow your mask outward; if you want that, you want a proper morphological dilation elsewhere. And as always in this pack: alpha software, so check the scale-factor math again after an update before trusting it in a coordinate-sensitive pipeline.