PM Image Loader

The core "Load Image" node is fine. It's also a blind dropdown of filenames, and when your input folder has a hundred files with names like img_0231.png, picking the right one is a guessing game. PM Image Loader is the PM Manager pack's answer: same job - pull an image out of ComfyUI's input folder and hand it to the graph as an IMAGE tensor - but with a picker that actually previews the file before you commit. If you live in img2img, inpainting, or ControlNet workflows, that preview saves you real time.

It's also just a good citizen under the hood. It hashes the file so that if you replace an image on disk, ComfyUI re-reads it instead of serving you a stale cached copy - a quiet fix for a classic "why is it still showing the old image" annoyance.

How it works

Read the source and you'll find it's core LoadImage, reimplemented with a bit more care. It handles multi-frame formats (GIFs, MPO) by loading every frame as a separate image in the batch, it respects EXIF orientation so your phone photos aren't sideways, and it does the standard alpha handling: if the image has an alpha channel, you get a mask output that's black where the image is transparent. No alpha? You get an empty 64×64 mask instead, which is what the rest of the ecosystem expects so downstream nodes don't choke.

Inputs and outputs

• image - a dropdown of image files in your ComfyUI/input folder. This is the only input, and the pack's UI also gives you an upload button so you can drop a file in without leaving the graph.
• image - the IMAGE output. Wire it into VAE Encode for img2img, a ControlNet preprocessor, an upscaler, or anything else that eats pixels.
• mask - the MASK output, from the image's alpha channel. Wire it into mask-aware nodes for inpainting.

Yes, the input and first output are both called image - the input is the file you're choosing, the output is the tensor. One's a string, one's an image; ComfyUI keeps them straight, and so will you.

Installing it

It ships with the whole PM Manager pack:

cd ComfyUI/custom_nodes
git clone https://github.com/pixel-magician/ComfyUI-PM-Manager.git
# restart ComfyUI

Or install via ComfyUI Manager by searching for ComfyUI-PM-Manager. No models to download, no special setup - the requirements (opencv-python, imageio-ffmpeg) are handled by the install.

Common issues

The dropdown only lists files in ComfyUI/input - if you can't see your file, it's not there, and the upload button is the fix. A "Invalid image file" error on load means the file vanished or got renamed between selection and execution; re-pick it. And the usual PM pack caveat applies: it's built on ComfyUI's newer extension API, so if the node doesn't appear at all, update ComfyUI rather than blaming the install.