SEGSPreview (CNET Image)

This is a debugging node, plain and simple, and it exists because ControlNet-inside-SEGS is exactly the kind of thing that fails silently. When you apply a ControlNet to your detected regions with ControlNetApply (SEGS) using a control_image, the pack crops that control image down to each detection's box and stashes the crop inside the SEGS. You never see those crops. If the crop is misaligned, upside down, or landed on the wrong region, your detail pass comes out wrong and you're left guessing why. SEGSPreview (CNET Image) just shows you the crops.

Reach for it whenever you're wiring a ControlNet into a detailing workflow and the results are off. Drop this node on the SEGS after ControlNetApply (SEGS), run once, and look at what the ControlNet is actually seeing per region.

How it works

The README is blunt about the scope: it "shows images configured with ControlNetApply (SEGS) for debugging purposes." It reads the control images that were cropped and attached to each SEG during the ControlNet apply step and renders them as a preview. That's the whole job. It doesn't change the SEGS, doesn't feed anything downstream - it's a window, not a valve.

One thing worth internalizing from the README: this preview only works for the control_image path. If instead you fed a segs_preprocessor into ControlNetApply (SEGS), the preprocessing happens on-the-fly during detailing and there's nothing pre-cropped to show here - you'd check those through the cnet_images output on the Detailer itself.

The inputs and outputs

There's exactly one input: segs - the SEGS that came out of ControlNetApply (SEGS). The output is an IMAGE list, one image per region's control crop, and because this is an output node the previews render right on the node in the ComfyUI graph. Nothing to configure; it either has control images to show or it doesn't.

How to install it

Via ComfyUI Manager: search ComfyUI Impact Pack, install, restart. Manually: cd ComfyUI/custom_nodes && git clone https://github.com/ltdrdata/ComfyUI-Impact-Pack, install the requirements into the same Python that runs ComfyUI (pip install -r requirements.txt, or ..\..\..\python_embeded\python.exe -m pip install -r requirements.txt on Windows portable), then restart. Automatic install was removed in v7.6, so the requirements step is on you. A small SAM model lands in ComfyUI/models/sams the first time the pack loads.

Common issues

The preview is empty. Almost always this means the SEGS never had control images attached - either you used the segs_preprocessor route (which this node can't preview, by design) instead of control_image, or the ControlNetApply (SEGS) node upstream didn't actually run. Confirm the apply node is in the graph and fed a control_image.

Nothing renders even though it ran. Remember SEGS before a detailer contains detection info, not necessarily images. This node specifically wants the ControlNet crops, which only exist after ControlNetApply (SEGS) - dropping it on raw detector output gives it nothing to draw. Wire it after the apply node, not after the detector.

That's about the extent of what can go wrong. It's a read-only inspection node; if the crops look right here, your ControlNet-in-SEGS setup is sound and you can move your attention to the detailer's denoise and strength instead.