Load Images (Original Size)

Load Images (Original Size) is the "stop hand-wiring fifty Load Image nodes" answer. Point it at a directory, and it hands back every image in it - PNG, JPG, JPEG, BMP, WEBP - as an IMAGE list, in alphabetical order, at their original dimensions. No resizing, no cropping, no color mangling: exactly what a dataset-prep workflow wants when the source images are already the size you need.

This is the kind of node that only makes sense once you've hit its use case, and then you can't live without it. Bulk processing a folder for a training run, running a style pass over a directory of references, feeding a whole sequence into a video pipeline - that's the job. It sits in the loaders half of the pack's "robust dataset preparation" pitch, and it pairs naturally with the pack's batch utilities (ImageListToBatch upstream of a sampler, or the folder metadata nodes if you're extracting prompts at the same time).

The mechanism is deliberately dumb, which is the point. It lists the directory, filters to the known extensions, sorts the filenames, then opens and converts each to RGB before normalizing to the float [0,1] tensor range ComfyUI works in. Two details matter: it returns a list of images (OUTPUT_IS_LIST), so ComfyUI can iterate the downstream graph per-image - the memory-friendly way to chew through a folder - and it re-runs whenever the folder changes, because its IS_CHANGED hash tracks file modification times rather than assuming the contents are static. If the path you typed doesn't exist, you get a clear "Directory cannot be found" error at validation time instead of a mid-run crash.

The usual gotchas: the directory input is a plain string path, not a file-picker - use an absolute path or you'll spend a while debugging. It only scans the top level; no recursion into subfolders. Sorting is alphabetical, which means frame_10.png loads before frame_2.png; zero-pad your numbered filenames if order matters to you. And RGB conversion silently drops any alpha channel - if you need transparency preserved, this isn't the node.

One honest caveat: it's a niche utility, and it's not magic. If your folder has 2,000 images at 4K, ComfyUI will happily try to hold all of them in VRAM as tensors, so think about whether you actually want the list approach or a per-image batch loop before pointing it at something huge.

Install is the standard AnotherUtils story: ComfyUI Manager, search "AnotherUtils", or cd ComfyUI/custom_nodes && git clone https://github.com/marcoc2/ComfyUI-AnotherUtils, then restart. No requirements, no weights, nothing extra to download - PIL and torch are already there.