Skeba Batch Images (Folder Loader)

The stock Load Image node is a single-file tool. Fine when you're testing one frame, a trap when you've got forty reference shots and you want the same workflow to run over all of them. Skeba Batch Images (Folder Loader) is the fix: point it at a folder, and it hands you every image in it as a real ComfyUI list that batch-aware nodes chew through one at a time.

It ships in the ComfyUI-H3-Reference-Library pack (also published as ComfyUI-Minimax-H3-Reference-Library), under Skeba AI Nodes - Utilities. It's one of the small utilities the pack consolidated from the older Skeba batching nodes, and it's the image side of the same idea as Skeba Batch Text (Prompt Loop) - feed a list in, let the workflow iterate.

How it works

Nothing clever under the hood, which is the point. The node lists the folder, filters by extension, sorts the files, then loads each one with PIL and converts to the same RGB float tensor format the core LoadImage node produces. The files come out as an IMAGE list, not one concatenated tensor, and it sets ComfyUI's OUTPUT_IS_LIST flag so downstream nodes know they're getting a batch and run once per image.

The three inputs that matter:

• folder_path - defaults to your ComfyUI input directory, so you can drop images in there and type a relative name. Absolute paths work too.
• image_extensions - comma-separated, defaults to png,jpg,jpeg,webp,bmp,gif. Add tiff or whatever your pipeline uses.
• sort_by - name, modified, or created. Pick modified if you keep overwriting files and want newest first.

The two optional ones are worth knowing: start_index skips the first N files, and max_images caps how many load (0 means all). Handy for spot-checking the first ten of a 500-frame folder before committing to the full run.

Outputs are images (the IMAGE list), filenames (a STRING list matching each image - great for labeling frames or writing a sidecar metadata file), and count (how many loaded, for driving a loop counter).

Where it fits

Wire images into any node that accepts an IMAGE list, or into Skeba Get Image From Batch when you want to select a specific frame by index instead of running everything. Pair it with the prompt-loop node and you've got a simple batch runner: all images, all prompts, one queue.

Installing it

ComfyUI Manager handles this pack - search "Skeba" or "H3-Reference-Library". Or clone it manually:

cd ComfyUI/custom_nodes
git clone https://github.com/nikaskeba/ComfyUI-H3-Reference-Library

Restart ComfyUI. There are no extra Python dependencies (the repo's pyproject.toml ships an empty dependencies = [] and there's no requirements.txt), so nothing fights your existing torch install.

Where people get burned

• A bad folder path fails silently. The node logs Folder not found to the console and returns an empty list - your downstream nodes just see zero images and no error. If nothing renders, check the console output first.
• name sort is lexical, not numeric. frame_10.jpg sorts before frame_2.jpg. Zero-pad your filenames (frame_02) or you'll run them out of order and wonder why the motion jumps.
• It all lands in VRAM. A huge folder loads every image at once. Use max_images or a subfolder for a first pass.