🖼️ Load Image Temporarily

ComfyUI's three directories have three different jobs: input/ is where loaders read from, output/ is where saves land, and temp/ is scratch that gets cleared. 🖼️ Load Image Temporarily is a Load Image node pointed at temp instead of input - upload a reference through it and the file goes to temp/, gets loaded, and is gone the next time you restart ComfyUI. Nothing permanent, nothing to clean up.

That's a real need, not a gimmick. If you iterate against a constant stream of throwaway reference images - a face you're testing a LoRA against, a screenshot someone sent you, a colour reference - your input/ folder turns into a museum of files you can't identify a week later. This node keeps the museum closed.

It works like the stock loader otherwise. image is a dropdown from the temp folder with an upload button wired to the temp folder, and the outputs are image (tensor), mask (alpha, inverted exactly like core Load Image, so transparent areas read as 1.0), width and height.

One small quality-of-life detail from the code: the dropdown deliberately starts empty rather than preselecting whatever temp file happens to exist first, so a freshly added node doesn't quietly load an unrelated image from a previous session. You pick the file or you get nothing.

The behaviour you need to design around

The temp folder is cleared on restart. That's the feature and the trap. Your workflow will load fine today and break tomorrow with a missing-file error, because the image it was pointing at no longer exists. If you're testing and iterating, fine. If you're about to share a workflow - or run it again next week - re-upload the image or switch to a loader that reads from input/, like the pack's own Load Image Advanced, which also extracts the positive prompt from the file's metadata while it's at it.

Temp images come from this node's upload, not from everywhere. Files you added through other loaders went to input/, and files that Preview Image nodes wrote are in temp but are outputs, not inputs. The two folders are genuinely separate, which is exactly the point but does surprise people looking for an image they know they "loaded".

No metadata recovery. This node outputs the image, a mask and dimensions - nothing else. If you want the embedded prompt too, that's the other loader.

The placeholder-mask thing

When the image has no alpha channel, core ComfyUI's loader pattern produces a small placeholder mask rather than a full-size black one, and this node follows that convention. Usually invisible - the standard masking nodes resize or ignore it. But if you've ever had a mask operation complain about mismatched sizes on an image you'd swear was the right dimensions, a missing alpha channel is a likely culprit.

Install

ComfyUI Manager → search ComfyUI-mnemic-nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MNeMoNiCuZ/ComfyUI-mnemic-nodes

Restart ComfyUI. No model downloads. Nothing special for this node - the pack's requirements install a stack of other packages (transformers, opencv-python, tiktoken, piexif…) for its other nodes, which is the usual grab-bag tax.