Image Loader From Path [LP]

Core ComfyUI's LoadImage node only knows how to open one specific file you pick from a dropdown of what's already in your input/ folder. This one is built for the batch case: point it at a folder - local or absolute path - and pull multiple images, in order, with filenames attached, instead of loading one image at a time by hand.

What it's for

This is the loader half of a batch-processing setup. If you're stepping through a directory of source images (product photos, dataset images, frames you've extracted elsewhere) and want ComfyUI to actually load a batch of them per run rather than you re-selecting one file at a time, this is the node. The README frames it plainly: load from a specific folder or path, either absolute or local (from the input folder), pull images sequentially by number as you cycle through, load them in batches, and get filenames back alongside the pixels.

That last bit matters more than it sounds - a lot of automation breaks down specifically because you lose track of which output came from which input once everything's flowing through a pipeline as anonymous tensors. Getting filenames back means you can carry that identity through to your save step.

Inputs and outputs

• input_folder - an enum populated from subfolders already present under ComfyUI's local input/ directory. Pick from what's there, or leave blank and use input_path instead if you want to point somewhere outside input/ entirely.
• start_index (INT, 0–9999, default 0) - where in the folder's file listing to start reading from. This is the "step" pointer for cycling through a folder across multiple queue runs - set it via an incrementing primitive if you're processing the folder one batch at a time across several queues.
• max_images (INT, 1–9999, default 1) - how many images to load in this batch, starting from start_index.
• white_bg - enum, disable/enable. Fills transparent areas with a white backdrop rather than leaving alpha intact - worth enabling if you're loading PNGs with transparency into a step downstream that expects a plain RGB image and chokes on alpha.
• patterns (STRING, default *.jpg|*.png|*.jpeg) - pipe-separated file extensions to match.
• input_path (optional, STRING) - an absolute or explicit path, for pointing this outside ComfyUI's local input/ folder entirely, overriding the input_folder dropdown.

Outputs: IMAGE (a list, not a single image), MASK (a list), Filename STRING (a list - the source filenames, matched position-for-position with the loaded images).

Installing it

ComfyUI Manager: search "ComfyUI-LevelPixel," listed as "Level Pixel," install and restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/LevelPixel/ComfyUI-LevelPixel.git, restart ComfyUI. No model downloads for this one - it's filesystem I/O.

Common issues

input_path, if used, needs to be reachable by the actual process running ComfyUI, not by whatever machine you're viewing the UI from - a common trap on remote or containerized ComfyUI setups where "the path on my desktop" isn't the same filesystem the server sees.

The outputs here are lists, and connecting a list output into a socket expecting a single item is a frequent first mistake - if max_images is greater than 1, downstream nodes need to be built to accept a batch, or you'll want to pair this with something that indexes into the list rather than assuming one image.

patterns is pipe-separated (*.jpg|*.png), not comma-separated - a very easy typo to make once and never notice since it fails silently by just matching nothing.