Icy LoadImage

LoadImage is the front door of most ComfyUI workflows - you drop in a reference image, and everything downstream (img2img, ControlNet, IP-Adapter, face swaps) hangs off what it outputs. Icy LoadImage is that same front door with a curtain over it: the thumbnail stays hidden until you hover or select the node. That's the whole point of the pack, and if you've ever shared your screen while your reference folder was full of things you'd rather not explain, you get it immediately.

IcyHider's author built the pack for people who "work with Comfy in public or do spicy content in their parents house," and this node is where that starts. On startup the pack subclasses ComfyUI's core node classes - same inputs, outputs, and behavior, new IcyHider Comfy Core category - and the extension's JavaScript covers or blurs their previews. So this is core LoadImage; the load logic is untouched.

What it does

From the schema, one input and two outputs:

• image - a dropdown of every file in your ComfyUI/input folder (and subfolders). Pick one.
• Output IMAGE - the image as a tensor, ready for VAEDecode output, samplers, ControlNet, or whatever.
• Output MASK - the image's alpha channel as a mask. If the image has no alpha, this is a fully-white (opaque) mask. For a reference image you usually ignore it; for compositing, it's the whole ballgame.

The node remembers your last pick, so you can iterate without re-selecting. Drop a new file into input/ and hit refresh in the node to see it in the list.

The privacy angle, made concrete

The value here isn't the loading - it's that the thumbnail doesn't show. In Avalanche Mode (on by default), the pack hides every node's content until selected; with Avalanche off, only Icy nodes hide on hover. Your reference images stay blurred or covered on screen, then snap back the moment you mouse over. Great for screenshots, great for screen-share, great for "no, that's a perfectly innocent image of a canyon, why do you ask."

Two honest caveats. First, the hiding is DOM-based, so on ComfyUI's newer Nodes 2.0 canvas frontend it can silently not engage - the legacy canvas is the pack's home turf. Second, the cover is screen-theater: it does nothing about the file sitting in your input folder, so if the person next to you has folder access, it's not hiding anything.

Install

No dependencies, no model downloads - it's a JS extension plus a thin wrapper:

# ComfyUI Manager: search "IcyHider", Install, restart

# or:
cd ComfyUI/custom_nodes
git clone https://github.com/icekiub-ai/ComfyUI-IcyHider.git
# restart ComfyUI

Then drop your images into ComfyUI/input as usual and pick them from the node. If you change Avalanche or hide mode, a page reload is required for it to take effect - that's a pack quirk, not a bug.