Gaussian Blur Mask (SEGS)
Feather every mask in a detection batch at once
- segs
- SEGS
Mask feathering, applied to a whole SEGS collection in one shot. If you've ever composited a refined patch back onto an image and seen a faint hard-edged rectangle where the new pixels meet the old ones, this is the fix. It blurs the edges of every mask inside a SEGS batch so the paste-back blends instead of snapping.
Where it sits in the pipeline
Impact Pack's core trick is detect → crop → resample at full resolution → paste back. That last step is where seams happen: if the mask defining "this is the region I replaced" has a hard edge, SEGSPaste composites a crisp boundary and you get a visible line, especially on faces where skin tone or lighting shifted slightly between the original and the reprocessed patch. Soften the mask edge before pasting and the transition disappears into a gradient instead of a cliff. This node does that softening for every element in a SEGS collection at once, so you don't have to touch each detection individually.
It's the SEGS-aware sibling of the pack's plain "Gaussian Blur Mask" node (which works on a single MASK). Use this one when you're operating on SEGS - right after a detector, or anywhere downstream in a SEGS pipeline - instead of converting to masks and back.
The inputs and output
segs(SEGS) - the detection batch whose masks you're softening.kernel_size(default 10, range 0–100) - the blur kernel diameter in pixels. Bigger kernel, wider spread of blur.sigma(default 10, range 0.1–100) - the Gaussian's spread/intensity. Higher sigma, softer and more gradual falloff.
Output is a single SEGS, same collection, masks blurred. It slots straight back into the rest of a SEGS pipeline - feed it to a Detailer, or straight to SEGSPaste if you're just tidying up masks before compositing.
Tuning it
Default values are a reasonable starting point; you rarely need to push either slider hard. A few pixels of blur is usually enough to kill a visible seam without eating into the detail you just spent compute generating. Go too far and you get the opposite problem - the mask edge softens so much that the refined region visibly smears into the surrounding, unrefined pixels, or the effective coverage shrinks so much the detail pass looks like it barely did anything. If a paste-back looks blurry or washed out at the edges rather than seamless, that's this node overdoing it, not the sampler underdoing it.
Installing it
Standard Impact Pack install - there's no separate dependency for this node specifically. Through ComfyUI Manager: search ComfyUI Impact Pack, Install, restart. By hand:
cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Impact-Pack
cd ComfyUI-Impact-Pack
pip install -r requirements.txt
(portable: python_embeded\python.exe -m pip install -r requirements.txt), restart. No models needed - this is pure image-processing math, not a detector or a sampler.
Common issues
Mostly it's a tuning problem, not a bug: too little blur and the seam is still visible; too much and detail bleeds or the effective mask shrinks to nothing. Start small and nudge up.
Worth knowing going in - this whole corner of the pack is genuinely under-the-radar. Most people who install Impact Pack are there for FaceDetailer's one-click behavior and never build a custom SEGS pipeline that needs manual mask feathering; this node is for when the default Detailer's own blending isn't cutting it and you're assembling the detect/refine/paste steps by hand. If you're at that point, you probably already know why - and if you're not, the built-in Detailer nodes handle basic feathering internally without you needing to reach for this.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| segs | SEGS | — | |
| kernel_size | INT | 100–100 | — |
| sigma | FLOAT | 10.00.1–100 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| SEGS | SEGS | — |