BatchImagesFromPath

You know the workflow: you've got a folder full of frames, crops, or reference images and you want all of them as one IMAGE batch - for an image-to-video model, a batch upscale, a comparison grid, whatever. Dragging them in one by one is misery, and ComfyUI's Load Image only does a single image. This node is the shortcut: point it at a directory, give it a pattern, and it loads everything that matches as a list of images.

It's part of ComfyUI-Przewodo-Utils, one of those "I wrote this for my own workflow and published it" utility nodes. Nothing fancy, but it closes a real gap.

How it works

The node does three things: resolves the path, globs for files, and converts each to a tensor.

• Path resolution. The path can be absolute, or relative to ComfyUI's output folder - the code runs it through ComfyUI's get_save_image_path, so a relative path like my_frames means <ComfyUI>/output/my_frames. If you give an absolute path that doesn't exist, the node will actually create the directory for you (that's a quirk of the implementation, not a feature to rely on).
• Pattern matching. A standard glob pattern. *.jpg, image_*.png, whatever. Leave it empty and it loads all image files in the folder. It globs non-recursively, so subdirectories are not searched - if you want recursive scanning, this isn't the node.
• Loading. Each file goes through PIL, gets EXIF_TRANSPOSEd (so phone photos load with the correct orientation, a detail most loader nodes skip), and is converted to a float tensor normalized to 0–1.

The output, named image, is a list of IMAGE tensors - one per file, in glob order (which is basically alphabetical by filename).

The inputs that matter

• path - directory to scan. Absolute, or relative to the output folder.
• pattern - glob pattern; empty means "everything."

That's it. Two text boxes, one output.

Installing it

From ComfyUI Manager, search ComfyUI-Przewodo-Utils. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/przewodo/ComfyUI-Przewodo-Utils.git

Restart ComfyUI. This node's dependencies are just PIL and torch - the pack's heavier requirements (teacache, gguf, wanblockswap, Rife-TensorRT) are for the video sampler and don't affect it.

Where people get burned

• The relative-path rule surprises everyone once. If you type C:/Users/you/frames you get your absolute path. If you type frames expecting it to look next to the workflow, you're looking at <ComfyUI>/output/frames instead. The node even tells you what path it resolved to in the console - read that line before hunting for files.
• Pattern typos silently return nothing. If nothing matches, the node returns None, and whatever you wired downstream will just fail with a confusing error. Start with the pattern empty to confirm the folder loads, then narrow it down.
• Memory. Every matched image lands in the batch at once. A folder of 4K frames will happily eat VRAM. Load only what you need with a tighter pattern, or you'll blame the wrong node when you OOM.