Load Input/Output Image

Stock ComfyUI's Load Image node only ever shows you what's sitting in input/. That's fine right up until you generate something, it lands in output/, and you want to immediately reuse it - feed it into an upscaler, run it through a color pass, chain it into another sampler. Now you're alt-tabbing to your file manager, copying the PNG into input/, and re-opening the dropdown. Load Input/Output Image exists to skip that step: it's a drop-in loader that browses both directories at once, subfolders included, and keeps the relative path intact when you pick a file.

What it actually does

It's the same job as the built-in loader - pick an image, get it back as a tensor - just pointed at a wider search space. Under the hood you're still just picking a file from a dropdown; the difference is the dropdown's scope. If your output/ folder has a 2026-08-04/ subdirectory ComfyUI made for you, this node walks into it rather than stopping at the top level, and the path it shows you (and stores in the workflow JSON) preserves that structure. That last part matters more than it sounds - it's what makes a saved workflow re-openable later and still pointing at the right file instead of a flattened filename that no longer resolves.

Practically, this node earns its keep in two spots: iterative workflows where you're chaining passes (generate → grade → upscale → generate again from the graded result) without manually shuffling files, and reviewing/reusing past outputs without leaving the graph.

Inputs and outputs

There's exactly one field to think about:

• image - a dropdown populated from both input/ and output/, subdirectories included. That's the entire required input; there's nothing else to configure.

Two outputs, both standard:

• image (IMAGE) - the loaded picture, wire it wherever an image would normally go.
• mask (MASK) - pulled from the image's alpha channel if it has one. If you loaded a flat JPEG with no transparency, expect a blank/full mask rather than anything meaningful - same behavior as the stock loader.

One thing worth knowing: this node is flagged as an output node internally, which in ComfyUI terms just means it'll always execute even if nothing downstream is connected to it - that's normally how ComfyUI decides which nodes to bother running. Practically it just means you can drop it in, pick an image to eyeball the thumbnail, and not worry about it getting pruned because you haven't wired anything else up yet.

Installing it

Two options, same as any custom node pack:

• ComfyUI Manager - search for ComfyUI-Light-N-Color (the pack title in Manager listings has a stray leading space, ignore it), install, restart.
• By hand:

  cd ComfyUI/custom_nodes
  git clone https://github.com/KAVVATARE/ComfyUI-Light-N-Color.git
  

  then restart ComfyUI. (Heads up: the README's own copy-paste install snippet still has the clone URL as github.com/yourusername/... - a template the author never swapped out. Use the real one above, or just let Manager handle it.)

No model downloads, no GPU-heavy dependencies - the whole pack leans on numpy, torch, opencv-python, and pillow, all of which ComfyUI already needs. If those import cleanly elsewhere in your install, this node will work.

Troubleshooting

• Empty or near-empty dropdown: this node can only show you what's actually in input/ or output/. A fresh ComfyUI install with no generations yet means an empty output/, so the dropdown will look sparse until you've run at least one job.
• A big output/ history: since it walks subdirectories, a folder with months of generations in dated subfolders means a longer list to scroll through. There's no search/filter described in the node itself, so if your output folder has grown large, it's worth knowing you're browsing all of it, not a filtered recent-files view.
• Missing/red node when opening someone else's workflow: this is a custom node, not core ComfyUI - if a shared workflow JSON references LoadInputOutputImage and the recipient doesn't have this pack installed, it'll show up broken. That's true of every node in this pack, not a bug specific to this one; point them at the install steps above.