Load Image From Path

ComfyUI's built-in LoadImage is great until it isn't. It only lets you pick from the input folder via a file browser, it returns no usable path, and it stares blankly at anything that isn't a plain image file. This node - Load Image From Path - is the fix for all three. You give it a path, it gives you the image, the mask, and the path itself as a data output.

The path-as-data part is the reason this node exists. The author says they built it specifically to feed workflows like ComfyUI_Qwen2-VL-Instruct, where a vision-language model wants to know the file path of the image it's reasoning about, not just the pixels. If you've ever tried to make a VLM tell you which image in a folder it's describing, you'll know why a loader that returns PATH instead of a silently-dropped string is worth having.

The inputs and outputs that matter

There's exactly one input: image, a plain STRING - the file path. No file browser. Paste it, or drag the file onto the field, and the node strips any quotes that Windows drag-and-drop sneaks in.

Four outputs:

• IMAGE - the pixels, ready for anything.
• MASK - an alpha mask. If the image has an alpha channel it becomes a mask (white = transparent, the standard ComfyUI convention); if there's no alpha you get a fixed empty 64×64 mask, same as stock LoadImage.
• STRING and PATH - both carry the file path as text, so you can pass it to a node that wants the raw string or one that's typed for PATH.

The Krita party trick

The headline feature is .kra support. A Krita file is secretly a zip archive, and this node opens it, finds mergedimage.png, and loads that - so you can feed an active painting file straight into img2img without exporting first. For anything else it falls through to standard loading. The mechanism is borrowed and adapted from JDC_ImageLoader in Jordach's comfy-plasma, and it's the kind of small quality-of-life thing that makes you wonder why ComfyUI doesn't do it natively.

Installing it

It ships in the ComfyUI-ARZUMATA grab-bag, so:

cd ComfyUI/custom_nodes
git clone https://github.com/ARZUMATA/ComfyUI-ARZUMATA

Restart ComfyUI, or install via ComfyUI Manager by searching ComfyUI-ARZUMATA. Zero extra dependencies, zero model downloads - it's PIL, numpy, and torch, all of which ComfyUI already has.

Where people get burned

• No file picker. If you're used to the stock LoadImage's nice browser, this feels backwards. The trade is that the path becomes data you can script and share.
• Krita files without mergedimage.png error out. Save-with-history or some layer setups won't include it; check the file.
• Typed paths are fragile. A missing file is caught at queue time with a "No file found" validation error - decent, but you'll still be hunting down the typo. On Windows, forward slashes work, so prefer those.
• The alpha mask is inverted from what a paint program shows you: the transparent area comes out as the mask's white region. That's deliberate (it matches ComfyUI's masking convention), but it surprises everyone once.

It's a small node in a small pack, and if you never load .kra files or hand paths to a VLM, you don't need it. If you do either, it's the one you'll reach for instead of the stock loader every single time.