🖼️ AGSoft Load Images From Dir

Stock ComfyUI loads one image at a time from a dropdown. When you need to batch-process a folder of images - a hundred frames from a video, a dataset you're captioning, a pile of reference shots - the standard workflow is drag a bunch of Load Image nodes and wire them all in. That's miserable. AGSoft Load Images From Dir loads every matching file from a folder in one shot, without resizing and without touching the alpha channel.

The node's own description spells out the guarantees: original size preserved, alpha preserved, and it supports natural sorting, ranges, and extension filters. If nothing matches your filter, it raises a clear error instead of crashing ComfyUI - a small thing that saves a lot of "why is my graph red" staring.

How it works

• folder_path - the directory to scan. Absolute or relative; the node doesn't care where it lives.
• filter_type - *.* for everything, or a specific pattern: *.png, *.jpg, *.jpeg, *.webp, *.bmp, *.gif. custom switches to the custom_filter field, where you can give it multiple patterns like *.png, *.jpg (a bare png gets normalized to *.png).
• start_index / end_index - a range into the sorted file list. Both default to 0, which means "from the start" / "to the end" respectively. Want files 5 through 10? Set 5 and 10. (0 at the end position = through the end of the list.)

Outputs: image as a list of IMAGE (one per file), a parallel filename list, the resolved folder_path, and number_of_files so downstream logic knows how many it's dealing with. Note the list outputs - this is what makes it work with list-driven batch nodes rather than expecting a single batched tensor.

Installation

From comfyui-AGSoft:

cd ComfyUI/custom_nodes
git clone https://github.com/Art-xmaster/comfyui-AGSoft.git
# restart ComfyUI

Or ComfyUI Manager → search "comfyui-AGSoft". numpy + opencv-python; no model downloads.

Where people get burned

The filename-vs-content mismatch: files are loaded in sorted order, so if you set a range assuming your own manual order, you may get different images than you expected. Sort your files or check number_of_files first. Also remember you're getting a list - if the node you're feeding wants a single batched IMAGE, you'll need a list-to-batch step. And this is a raw loader: no auto-cropping, no resize, no EXIF rotation. Feed it what you want to process. For one folder at a time it's genuinely the fastest way to get a pile of images into a graph.