Nodes/ComfyUI-IcyHider/Icy LoadImageOutput
ComfyUI Node

Icy LoadImageOutput

Load From the Output Folder and Iterate Faster

By icekiub-ai·Created 9 months ago·Updated 9 months ago· 31
Icy LoadImageOutput
    • IMAGE
    • MASK
    image

    IcyLoadImageOutput is LoadImage's cousin that reads from your output folder instead of your input folder. The use case is feedback loops and iteration: you generate, grab the result right back off disk, and feed it into another pass - upscale, img2img again, face-fix, whatever - without copying files around. It's the node that makes "generate → refine → refine again" a loop instead of a chore.

    The "Icy" prefix is the pack's privacy layer, not a different loader. IcyHider subclasses core ComfyUI nodes at startup - same inputs, outputs, and behavior, moved to the IcyHider Comfy Core category - and its JavaScript covers previews until hover. This is core LoadImageOutput with a coat on, which is genuinely the point here: when you're feeding your own outputs back in, the node's thumbnail is basically your last render, and that's exactly what you may not want staring back at you on a shared screen.

    What it does

    From the schema, one input, two outputs:

    • image - a dropdown of files in your ComfyUI/output folder. The author's own description calls out the useful bit: click the refresh button and the node updates the list and automatically selects the first image - the newest output - which is the whole iteration trick. Generate, hit refresh, and you're already looking at your latest render without hunting for it.
    • Output IMAGE - the loaded image as a tensor.
    • Output MASK - the alpha channel as a mask, if present.

    Wire the IMAGE into a VAE Encode for an img2img pass, an upscaler, a ControlNet, an IP-Adapter, or just a comparison node. It behaves like any image source.

    Why you'd actually reach for it

    The two-pass workflow is the classic: generate at native res, then reload that exact output for a detail or upscale pass at low denoise. Loading from the output folder means your "source" image is always your latest generation - no manual file moves, no stale copies. Combined with the auto-select-on-refresh behavior, this turns a two-node chore into a one-click loop. It's also how people build "polish the last output" chains that are easy to read on the graph.

    One honest caution: it reads from disk, so the image must be saved before it can be loaded back. If you hook this into a workflow that hasn't actually written anything yet, you'll get a "no image selected" moment until something downstream produced a file. That's expected, not a bug.

    Install

    Same frontend-only pack, zero dependencies:

    # ComfyUI Manager: search "IcyHider", Install, restart
    
    # or:
    cd ComfyUI/custom_nodes
    git clone https://github.com/icekiub-ai/ComfyUI-IcyHider.git
    # restart ComfyUI
    

    Pack-wide: the hiding is DOM-based, so the newer Nodes 2.0 canvas frontend may not engage it, and the cover is screen-only - the file in your output folder is still there for anyone with access.

    CategoryIcyHider Comfy Core

    Inputs (1)

    NameTypeDefaultDescription
    imageCOMBO

    Outputs (2)

    NameTypeDescription
    IMAGEIMAGE
    MASKMASK