Nodes/Nukun_ComfyUI_Nodes/Split Masks (Nukun)
ComfyUI Node

Split Masks (Nukun)

Split masks without the conditioning, for when that's all you need

By OnekoSL·Created 3 months ago·Updated 11 days ago· 1
Split Masks (Nukun)
    • mask_1
    • mask_2
    • mask_3
    width1024
    height1024
    region_count2
    orientation
    split_10.50
    split_20.67
    overlap0.00

    The simplest node in the regional family, and that's its whole appeal. Split Masks (Nukun) takes a width, height, and an orientation, and builds two or three split masks - nothing else. No conditioning, no encoding, no model patching. You get mask_1, mask_2, mask_3 and you wire them into whatever consumes masks in your own graph: RES4LYF regional conditioning, native ConditioningSetMask chains, or a custom workflow that wants raw geometry.

    You'll see the same split geometry in the conditioning nodes of this pack, but those bundle mask-plus-apply. This one exists for the case where you already have your conditioning handled and just need the masks - and, honestly, for the case where you want to see the masks before committing to a regional workflow at all. Preview these, decide the layout is right, then rebuild with the bundled version.

    How it works

    Inputs: width, height, region_count (2 or 3), orientation (horizontal = left/right split, vertical = top/bottom), split_1 (default 0.5), split_2 (default 0.67, only used at 3 regions), and overlap (0–0.25). At region_count 2, split_2 is ignored and mask_3 is an empty mask.

    overlap is the one input worth spending time on. At 0 the bands meet at a hard line; raise it to 0.1–0.2 and neighboring regions expand across the boundary with a soft blend. If you're feeding these to a ConditioningSetMask-style consumer, hard edges can leave a visible seam in the output - a small overlap usually hides it. The masks are pure tensors, so previewing them with a standard mask-preview node is instant; there's no reason not to look before you wire anything downstream.

    Installing it

    Same pack, standard install:

    cd ComfyUI/custom_nodes
    git clone https://github.com/OnekoSL/Nukun_ComfyUI_Nodes.git
    

    Restart ComfyUI, or use Manager (search "Nukun"). No dependencies - it's a handful of tensor operations, which is exactly why it behaves predictably.

    Common issues

    The orientation naming bites everyone once: horizontal produces a vertical boundary line (left/right regions), vertical produces a horizontal boundary (top/bottom). If your split looks flipped, that's it. And remember what this node does not do: it won't apply the masks to anything, and it won't tell you which prompt belongs to which region. If you connect masks and see an unchanged image, that's expected - the masks are just inputs to a consumer you need to wire yourself. The bundled Native Regional Split Conditioning is the version that does the whole job in one step; reach for it if this feels like too much wiring.

    CategoryNukun/Mask

    Inputs (7)

    NameTypeDefaultDescription
    widthINT10248–16384
    heightINT10248–16384
    region_countINT22–3
    orientationCOMBO2 options: horizontal, vertical
    split_1FLOAT0.500.01–0.99
    split_2FLOAT0.670.01–0.99
    overlapFLOAT0.000–0.25

    Outputs (3)

    NameTypeDescription
    mask_1MASK
    mask_2MASK
    mask_3MASK