LoadImagePath

The first thing that trips people up on ComfyUI-Hallo is that you can't reach for the default LoadImage node. HalloNode doesn't want a tensor sitting in GPU memory - it wants a file path on disk, because the actual Hallo inference runs as a separate Python script that reads the image from a path. LoadImagePath exists to hand it that path.

What it does

On load, the node scans your ComfyUI input/ folder, filters for bmp, jpg, png, webp, and jpeg, and presents the results as a dropdown. Pick one and it resolves and returns the full annotated path, ready to feed straight into HalloNode's source_image input. It also sets the upload flag, so you can drag-and-drop a fresh image right onto the node and ComfyUI copies it into input/ for you, same as the core loaders.

The whole workflow for this pack is therefore:

LoadImagePath ──► HalloNode.source_image
LoadAudioPath ──► HalloNode.driving_audio
HalloNode.Video ──► PreViewVideo

Two gotchas worth knowing

• The list is built when the node loads. Drop a new image into input/ while ComfyUI is running and it won't show up in the dropdown until you reload the workflow (or restart). Annoying, but harmless.
• It doesn't look in subfolders. Where the core LoadImage walks the whole input tree, this node does a flat os.listdir of the top-level input directory. Those images ComfyUI stashed in input/whatever/ for you? Invisible to this node. Move the file to the root of input/ and it appears.

Install

This is part of the AIFSH/ComfyUI-Hallo pack, so you get it by installing the pack: ComfyUI Manager → search ComfyUI-Hallo, or git clone https://github.com/AIFSH/ComfyUI-Hallo.git into ComfyUI/custom_nodes/ plus pip install -r requirements.txt. Same install as HalloNode - the loader is just the plumbing in front of it. You don't even need to think about this node on its own; it's the thing you wire before the thing that does the actual work.