Nodes/ComfyUI Inspire Pack/HED Preprocessor Provider (SEGS)
ComfyUI Node Runs on cloud

HED Preprocessor Provider (SEGS)

Soft-edge ControlNet for detected regions

By ltdrdata·Created 3 years ago·Updated 9 months ago· 805
HED Preprocessor Provider (SEGS)
    • SEGS_PREPROCESSOR
    safetrue

    This is the soft-edge sibling of the SEGS preprocessor providers, and it only earns its place if you're using Impact Pack's SEGS pipeline. Quick recap: SEGS is Impact Pack's batch-of-detected-regions format, and its detailer nodes resample each region for quality. You can steer that per-region refinement with a ControlNet, but the SEGS system needs the preprocessor packaged its own way. This node hands you exactly that - a HED soft-edge preprocessor wrapped as a SEGS_PREPROCESSOR.

    HED (Holistically-nested Edge Detection) is the soft-edge preprocessor - nowadays often just called "softedge." As our ControlNet notes put it, it produces gradient edges with varying thickness and is more forgiving than Canny, which makes it a good fit for organic subjects like faces and bodies where you don't want to lock every hard contour. That forgiveness is precisely why it pairs well with region detailing: you get structural guidance without fighting the resample.

    How it works

    Like the other providers here, this node doesn't preprocess anything the moment it runs. It's a provider - it emits a SEGS_PREPROCESSOR object that says "use HED soft edge." You feed that into an Impact Pack node that applies ControlNet to SEGS, and the actual edge extraction happens per detected region at execution time, driving a soft-edge ControlNet that guides each region's refinement. So instead of running one HED pass over the whole image, you're controlling only the cropped regions the detailer is working on.

    The inputs and outputs

    • safe (BOOLEAN, default true) - the standard safe-mode flag carried over from the ControlNet aux preprocessors; it keeps the preprocessed output in a conservative range to reduce artifacts. Sensible to leave on.
    • Output is a single SEGS_PREPROCESSOR, which goes into an Impact Pack Apply ControlNet (SEGS) node.

    That's the entire surface. The node is intentionally minimal - the work lands where you apply the result.

    How to install it

    ComfyUI Manager: Install Custom Nodes → search ComfyUI Inspire Pack → install → restart. Or from a terminal:

    cd ComfyUI/custom_nodes
    git clone https://github.com/ltdrdata/ComfyUI-Inspire-Pack
    

    Two dependencies are mandatory for this one. You need the Impact Pack (also by Dr.Lt.Data) for the SEGS system, and Fannovel16's comfyui_controlnet_aux for the HED preprocessor itself - the Inspire README names that pack as the source of the preprocessor code. Install both via Manager and restart. You'll also want a soft-edge / HED ControlNet model matching your base architecture on the applying node.

    Common issues

    If it does nothing at generation time, check the dependencies first: no comfyui_controlnet_aux means no HED preprocessor to run, and no Impact Pack means no SEGS node to apply it in. Because it's a provider, its output only connects to a SEGS ControlNet-apply node - nothing else understands a SEGS_PREPROCESSOR. And keep the ControlNet model's architecture aligned with your checkpoint; a mismatched softedge model is a common reason the control seems to have "little or no effect." If you want crisp, hard structural lines instead of soft gradients, this is the wrong preprocessor - reach for the Canny provider.

    CategoryInspirePack/SEGS/ControlNet

    Inputs (1)

    NameTypeDefaultDescription
    safeBOOLEANtrue

    Outputs (1)

    NameTypeDescription
    SEGS_PREPROCESSORSEGS_PREPROCESSOR