Output Image To Input

You just generated something in ComfyUI, it landed in your output/ folder, and now you want to feed that exact file back into another part of the graph - maybe to upscale it, maybe to use it as a reference for the next generation, maybe to batch-process a folder of results you already made. Normally that means digging through your file browser and re-uploading. Output Image To Input skips that: it lets you pick a file straight out of ComfyUI's output directory and hands you back a value you can wire into anything downstream.

Mechanically it's a dropdown node, the same pattern ComfyUI's built-in Load Image node uses, except instead of listing your input/ folder it lists output/. The one required field, image, is an enum populated at load time from whatever's actually sitting in that folder. If you've never generated anything yet, or the node hasn't refreshed its list, all you'll see is a placeholder entry telling you images will show up there once you have some - that's expected, not a bug, refresh the node (or the page) after your first run.

Here's the part that trips people up: the output isn't an IMAGE tensor, it's a STRING. So you can't wire it directly into something expecting pixels - you're getting a filename/path reference, not the decoded image. That makes sense once you see what it's built to pair with: this same pack ships Load Image With Name, whose required input is a plain string filename and whose outputs are IMAGE and MASK. Chain them - Output Image To Input picks the file, Load Image With Name actually loads it - and you've got a two-node "grab a past result and load it back in" combo, which is a nicer workflow than round-tripping through your OS file picker every time you want to iterate on an output.

Installing it: the pack this ships in, comfyui-mask-util, is from the same author (longgui0318) who built comfyui-magic-clothing, a ComfyUI wrapper around a virtual-try-on diffusion pipeline that had a brief moment of Reddit attention in 2024 and has been quiet since. This mask-util pack reads like the utility scraps from that project, generalized and split out - small plumbing nodes rather than anything with a big following, and worth knowing that going in so you're not hunting for docs that don't exist. And they largely don't: the GitHub README is literally just the title, no install notes, no usage examples. So install it the standard way - either open ComfyUI Manager, search "comfyui-mask-util", and hit install, or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/longgui0318/comfyui-mask-util

Restart ComfyUI afterward. There's nothing to download beyond the code - this is a pure file-listing node, no model weights, no extra Python packages beyond what ComfyUI already ships.

Where it bites: the dropdown is a snapshot taken when the node loads, not a live folder watch. Generate a new image after the graph is already open and this node won't see it until you refresh (ComfyUI Manager's refresh button, or reload the page). If your output folder has thousands of files, expect the dropdown to get unwieldy - this node has no search/filter, it's a flat list. And because the output is a string rather than an image, if you plug it somewhere expecting IMAGE you'll get an immediate type-mismatch error at graph-build time, which is ComfyUI's way of telling you that you skipped the Load Image With Name step.