JDCN_BatchImageLoadFromDir

JDCN_BatchImageLoadFromDir loads a folder of images as a batch - skipping the first N if you want, capping the total, and handing you the actual IMAGE tensors plus their names and paths. Where the file-list nodes just hand you paths, this one is where the bytes finally enter your graph. It's the natural endpoint of AnyFileList → loader, and it's built for the "load a frame sequence and process it" pattern.

How it works

Three inputs:

• Directory - the folder to scan.
• Load_Cap - how many images to load (default 1).
• Skip_Frame - skip the first N images before loading (default 0).

Under the hood it walks the folder and grabs every file matching the pack's giant built-in image extension list (jpg, png, webp, tiff, raw, the lot), converts each to RGB, and returns a list of IMAGE tensors. Then it slices by Skip_Frame and Load_Cap: Load_Cap = 0-ish edge case aside, you get file_paths[skip : skip + load].

Six outputs, but you'll mostly touch three:

• Images - the loaded tensors (a list of IMAGE).
• Image_Names - filenames without extensions.
• Image_Paths - the full paths of what actually loaded.
• Load_Cap, Skip_Frame, Count - echoes and the final count.

The two gotchas that matter

Order is not sorted. The folder walk uses OS order, same as AnyFileList. For a frame sequence frame1, frame2, ... frame10, that's likely not the order you'd expect on Linux. Zero-pad your filenames so OS order matches numeric order, or you'll be feeding shuffled frames into your pipeline.

Each image is its own tensor in a list. Downstream, whatever consumes Images needs to handle a list of IMAGEs, not one batched IMAGE. If your processor wants a single batched tensor, you'll need a batch/stack step in between. If the node's default Load_Cap = 1 surprises you by loading one image - that's the default, raise it.

When you'd reach for it

Frame-sequence batch img2img, batch upscaling, or any "process every frame of this folder" job. Pair the skip math with JDCN_BatchCounterAdvance's SkipFrame output for chunked processing with overlap, which is exactly what that counter was designed to feed.

Installing it

Part of ComfyUI-JDCN:

• ComfyUI Manager → Install Custom Node → search JDCN → install ComfyUI-JDCN → restart.
• Or:

cd ComfyUI/custom_nodes
git clone https://github.com/daxcay/ComfyUI-JDCN.git
cd ComfyUI-JDCN
pip install -r requirements.txt

Restart; under 🔵 JDCN 🔵. The pack depends on piexif; image loading uses Pillow, which ComfyUI already ships.

Common issues

Bad Directory → empty everything (watch the console log). Images that fail to open (corrupt files, weird formats) are skipped with a console error rather than crashing the batch - good, but it means Count may be lower than the folder's file count. And if colors look wrong, remember everything is force-converted to RGB; transparent PNGs and CMYK files get flattened. If you need a specific slice of a huge sequence, raise Skip_Frame rather than pointing it at the whole folder.