📦 Load Image Batch (OreX)

Loading one image at a time is fine until you have a folder of 500 and a batch job to run on all of them. Load Image Batch (OreX) is the pack's folder-walker: point it at a directory, give it a file pattern, and it serves images one at a time (or by index), cycling through the whole folder so each queue run processes the next file. The details are where it shows its maturity - the file list is frozen per run so it never picks up its own freshly-written outputs mid-batch, which is the classic infinite loop that eats your hard drive.

How it works

Give it folder_path (absolute, or relative to the input folder) and file_pattern (a glob like *.png or *upscale*). It scans the folder once per run, sorts the matches, and hands back one image per execution in mode:

• single_image - returns the file at start_index. Deterministic, good for testing.
• incremental_image (default) - returns the next file each time the queue advances, wrapping around to the start after the last one. This is the batch-workhorse mode.

Two safeguards matter. First, the file list is cached per label - once a batch named "Batch 001" is locked in, it keeps its original list even if files appear or vanish mid-run. That's what stops the snowball effect where every processed output gets added to the input pile. Second, the README is explicit: if batch_count / the folder is smaller than expected, files are not repeated just to fill a count - it stops rather than double-process.

Outputs are the useful quartet: image, filename, folder_path, and the metadata pair number_of_files / current_index (so downstream nodes know where you are in the run). There's also allow_rgba_output if you genuinely need alpha through the pipeline - otherwise everything is converted to RGB, which is the safe default for most samplers.

The inputs that matter

• mode - single_image vs incremental_image.
• folder_path + file_pattern - where and what to scan.
• label - a name for the batch; this is what anchors the frozen file list.
• seed - in incremental mode, changing the seed lets you re-run the same queue position; with a fixed seed the batch won't restart.

Install

ComfyUI Manager (search "comfyui-OreX") or:

cd ComfyUI/custom_nodes
git clone https://github.com/orex2121/comfyui-OreX

Restart. No extra dependencies.

Where people get tripped up

The seed/rerun behavior confuses people: the README stresses that files always process low-to-high and that seed only exists so you can restart the sequence - with fixed seed, re-running won't start over. Pair it with the pack's Save Image (create_processed_folder on) and your processed outputs go to a processed/ subfolder so the batch never re-reads them - that's the intended workflow, and it's why the two nodes are designed to work together.