Nodes/ComfyUI-Selective-Sigma-Detailer/Selective Sigma Detailer (Debug Preview)
ComfyUI Node

Selective Sigma Detailer (Debug Preview)

Finally see what the mask is masking

By Capacap·Created 4 months ago·Updated 4 months ago· 1
Selective Sigma Detailer (Debug Preview)
  • mask_ref
  • latent
  • latent
  • image

The whole point of the Debug Preview node is that the selective part of Selective Sigma Detailer is invisible unless you look. You run a generation, the mask is computed mid-sampling, and without this node you just get the result and have to trust it. This node turns that hidden mask into an image you can stare at, which is exactly what you want when a detailer is doing something unexpected - or when you're tuning coverage and ema and want to see the mask move.

How it works

It pairs with the Debug sampler's mask_ref output (type SSD_MASK_REF). You feed it two things:

  • mask_ref - straight from Selective Sigma Detailer (Debug).
  • latent - a passthrough. This is the sneaky bit: because the node returns the latent unchanged, you wire the latent from your sampler's output through this node, which forces ComfyUI to execute it only after sampling finishes. Without the passthrough, the mask wouldn't exist yet when the node ran.

Outputs are latent (the untouched passthrough, so it can sit inline in your graph without breaking anything) and image - the rendered mask as a standard IMAGE tensor. The render is the first latent channel upscaled 8x with nearest-neighbor, deliberately, so you see the latent-grid structure honestly instead of bilinear-smoothed mush. The mask also gets written to a temp PNG (ssd_mask_*.png) and shown in the UI, and the node is marked as an output node so the image is easy to find.

The image output is there for real composition, not just eyeballing - you can wire it into a comparison grid or save it alongside your renders, the way the pack's own README coverage sweeps show masks under each generated image.

What you'll actually learn from it

The README's sweep images are the best guide to what a healthy mask looks like: on a portrait it concentrates on the face, freckles, and hair, with the sky getting a visibly weaker but non-zero treatment. If your mask is a flat block or covers the whole frame at mid-coverage, your image has detail everywhere (unusual) or something's off. If it's near-empty, lower coverage isn't the fix - check that the detail pass is even running via the stats line in the console. Keep in mind the mask is a soft weighting, not a binary gate: subtle changes outside the bright regions are expected, because smooth areas receive an attenuated shift rather than full protection.

Install

Same pack as the other two nodes - Manager search Selective Sigma Detailer, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Capacap/ComfyUI-Selective-Sigma-Detailer

Restart ComfyUI. No extra dependencies, no model files, MIT licensed. One caveat: if you wire it up and the preview shows an all-black image, you've likely short-circuited the debug sampler (e.g. strength = 0), in which case no mask was ever captured - the debug nodes still populate a placeholder so the graph runs, but there's nothing real to show.

Categorysampling/custom_sampling/samplers

Inputs (2)

NameTypeDefaultDescription
mask_refSSD_MASK_REF
latentLATENT

Outputs (2)

NameTypeDescription
latentLATENT
imageIMAGE