🖼️ Load Image Advanced

Stock LoadImage gives you an image and a mask, full stop. Load Image Advanced does the same job but hands back the two things you actually want when you're pulling an existing image back into a graph: its file path and the positive prompt buried in its metadata, plus width and height so you're not chaining a separate size-detection node afterward.

How it works

This leans on the same trick ComfyUI uses to make workflow sharing possible in the first place - a ComfyUI output image carries the full graph that produced it embedded inside the file itself, so a single PNG is enough to reproduce or hand off a generation. This node reads that embedded data back out for you and surfaces the positive prompt specifically, alongside the file path so you know exactly which image on disk you're looking at.

The inputs and outputs that matter

• image - the file picker, same as stock LoadImage.

That's the only input. The value is in what comes back out:

• image / mask - the loaded image and its alpha channel, same as LoadImage.
• image_path - the full file path of the loaded image.
• positive_prompt - whatever positive prompt was extracted from the image's embedded metadata.
• width / height - the image's dimensions, no separate node needed.

How to install it

Via ComfyUI Manager: search ComfyUI-mnemic-nodes, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/MNeMoNiCuZ/ComfyUI-mnemic-nodes

Restart ComfyUI. No model downloads, no extra dependencies - it reads metadata that's already sitting in the file.

Common issues & troubleshooting

positive_prompt comes back empty. This node reads metadata that's genuinely embedded in the file - it doesn't invent a prompt if there isn't one to find. A re-exported, re-compressed, or format-converted image (a JPEG or WebP saved from a screenshot tool, for instance) has usually lost that embedded data entirely, and there's nothing this node can do to recover text that isn't there.

mask comes back blank. Same story as the stock LoadImage node's alpha mask behavior - if the source image has no alpha channel, you get an empty mask rather than an error. That's expected, not a bug.

Why bother over stock LoadImage? Mainly when you're doing anything metadata-aware: rebuilding a prompt from an old render without opening it in a separate viewer, driving a downstream text node straight from a loaded image's own history, or just skipping a couple of extra nodes (a size-detect node, a metadata-reader node) you'd otherwise need to bolt on. If you don't need any of that, stock LoadImage is perfectly fine and marginally simpler.