👺 Load Image By Index

ComfyUI's built-in Load Image makes you pick a file from a dropdown. That's fine for a few images and maddening for hundreds - you can't drive a batch pipeline from a dropdown. Load Image By Index inverts the model: you point it at a directory, and it picks an image by index, where the index comes from a seed input you can wire to anything. Programmatic image loading, finally.

How it works

Inputs:

• image_directory - the folder to scan.
• seed - used as the index, not a random seed. This is the key insight.
• sort_mode - filename (default), creation_time, modification_time, or size, which decides what "index N" means.
• reverse_sort - flips the order.

The node lists the directory, filters to common image extensions (png, jpg, jpeg, webp, bmp, gif - case-insensitive, notably more generous than the pack's counter node), sorts per sort_mode, then picks seed % len(files). That modulo is important: out-of-range indexes wrap around rather than erroring, so seed 999 on a 10-image folder loads image 9. Outputs are the image tensor, the filename (basename), and the filepath (full path) - so you get the identity of what you loaded, which is exactly what you want when a workflow is iterating over a folder.

The tricks worth knowing

It's an index, not a randomizer. Same seed, same sorted list, same image, every time. To walk through a folder sequentially, feed seeds 0, 1, 2… from a counter or 👺 List of Ints. To randomize which image loads, you need a seed source that varies - the pack's 👺 User Input (Seed) is built for exactly that.

Sorting changes what a seed means. If you're loading "image 5" you must know the sort order to predict which file that is. Filename sort is stable and predictable; creation-time sort is OS-dependent and can reorder if files are copied. The IS_CHANGED hook in the source only keys on the seed, so changing just sort_mode or reverse_sort may not trigger a re-execute - if you change the sort and the preview doesn't update, nudge the seed.

The folder must exist. Missing directory → hard error. And if the folder is empty or has no supported images, it errors too - there's no graceful "no files" path. Filter first with 👺 Count Images In Directory if you want a workflow that survives empty folders.

Performance. It rescans and re-sorts the directory on every execution. Fine for hundreds of files, wasteful if you're re-loading the same index repeatedly in a loop - nothing is cached. Also, it loads the whole image into memory each call; don't feed it a folder of 4K raws if you only need thumbnails.

Install

ComfyUI Manager (search "Nilor Nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/nilor-corp/nilor-nodes
cd nilor-nodes && pip install -r requirements.txt

Restart ComfyUI; it's under Nilor Nodes 👺 → IO. Needs only PIL/torch.

Bottom line

The right answer to "I want a workflow that picks images from a folder by number." Pair the seed input with any index source and you have a loop, a random sampler, or a folder walker in one node. Just keep the modulo-wrap and the re-scan cost in mind, and it'll quietly replace a hundred dropdown clicks.