Nodes/RC Image Compositor/RC Load Image (Alpha)
ComfyUI Node

RC Load Image (Alpha)

The loader that keeps transparency alive

By kj863257·Created 11 months ago·Updated 10 months ago· 16
RC Load Image (Alpha)
    • IMAGE
    • MASK
    image
    RGBA_modetrue

    ComfyUI's built-in LoadImage quietly flattens transparency in a way that bites anyone doing compositing: the RGB channels get baked, the alpha gets dumped to a mask output, and by the time you're compositing a cutout you've lost the thing that made it a cutout. RC_LoadImageWithAlpha is the pack's fix - it loads a PNG as RGBA, keeping the alpha channel live in the IMAGE tensor so your blend and layer-effect nodes actually see it. It's the "start a compositing workflow correctly" node in the RC Image Compositor suite.

    How it works

    It reads the file with PIL, converts to RGBA (preserving the existing alpha), and outputs two things:

    • IMAGE - a [1, H, W, 4] RGBA tensor. This is the important output: 4 channels means downstream nodes that understand alpha (all the RC compositing and layer-effect nodes) can blend it properly.
    • MASK - the alpha channel in mask form. Note the convention: it's 1 - alpha, so white = transparent areas, matching ComfyUI's built-in LoadImage behavior (white = where an action applies). If you feed this into an inpainting node, it'll target the transparent regions - which is usually exactly what you want for "fill in what's missing."

    The one optional input, RGBA_mode (default on), forces a 4-channel output even for images that have no alpha - adding an opaque channel instead of leaving it as RGB. Turn it off only if you genuinely need the image to stay 3-channel.

    It's a plain file-picker input, so you select your PNG from ComfyUI's input directory like any loader.

    Installing it

    ComfyUI Manager → search "RC Image Compositor" → install, restart. Manual:

    cd ComfyUI/custom_nodes
    git clone https://github.com/kj863257/ComfyUI_RC_Image_Compositor
    cd ComfyUI_RC_Image_Compositor
    pip install -r requirements.txt
    

    Dependencies: pillow, numpy, opencv-python. Nothing to download.

    Where people get burned

    • The MASK output is the transparency, not the opacity. It's 1 - alpha, so a fully transparent background reads as a white mask. If you expected the mask to highlight your subject, it won't - invert it (or just use the IMAGE output's alpha).
    • If your downstream node ignores alpha, this loader doesn't save you. Most non-RC nodes silently drop the 4th channel when they process an RGBA tensor. The whole point is to pair this with nodes that respect alpha - this same pack's compositor and layer effects.
    • RGBA_mode off + RGB image = no alpha to work with. You asked for RGB, you get RGB. It's a mode for pipelines that choke on 4-channel tensors, not for cutout work.

    This is the quiet hero of the compositing side of the pack. If you're bringing in PNG cutouts - from BiRefNet, from a background-removal workflow, from a sticker pack - load them with this node and everything downstream that respects alpha starts working the way it should.

    CategoryRC/Image

    Inputs (2)

    NameTypeDefaultDescription
    imageCOMBO1 options: example.png
    RGBA_modeoptBOOLEANtrueForce RGBA output. If no alpha, add opaque channel.

    Outputs (2)

    NameTypeDescription
    IMAGEIMAGE
    MASKMASK