🖼️ Load Image Advanced

🖼️ Load Image Advanced is the pack's beefed-up version of ComfyUI's stock Load Image. Same picking-a-file-from-input/ starting point, but it hands you six outputs instead of two, and one of them is the thing people actually want: the positive prompt that was used to make the image, dug out of the file's metadata.

That's a bigger deal than it sounds, because in ComfyUI the metadata is the payload. A PNG from SaveImage carries the whole workflow graph plus the execution prompt as text chunks in the file. Open a good image from CivitAI or a Reddit workflow share, drop it in this node, and you have the prompt back as a string you can wire into a text encoder. No retyping, no squinting at a screenshot.

The other outputs: image (the tensor), mask (the alpha channel if the image has one), image_path (the full path as a string), plus width and height as separate integers. That last pair is quietly useful - feed dimensions straight into a size selector or a latent node instead of typing them, and your img2img pass can't disagree with your source.

The catch: metadata is fragile

Here's the honest part, and it's worth understanding before you build a workflow on this output. Metadata survives exactly as long as nobody re-encodes the file:

• Reddit, Twitter/X, Discord inline previews and most CDNs recompress your upload. The text chunks are gone. This is why people are constantly surprised that a shared PNG won't behave.
• Any JPEG or WebP conversion drops it. Those formats have no PNG text chunks at all. A flat A1111 parameter string can sometimes survive in EXIF, but the ComfyUI graph does not travel.
• Screenshots never had it. Neither did files saved by an instance running with metadata disabled.

So positive_prompt is a string that is often empty, and it fails silently when it is. If you wire it into an encoder and get a blank generation, check the metadata before you debug the sampler - that's the failure mode here. Long-term, keep the original PNG of anything you care about; every copy you post is a derivative that's losing features.

Where it fits

Three real uses. First, prompt archaeology on images you downloaded - see what made them, then remix. Second, dimension plumbing: width and height out to whatever needs to know the canvas size, so the whole graph follows the loaded image. Third, inpainting setups: the mask output is the image's own alpha channel, so a transparent PNG becomes a ready-made mask without a segmentation pass, and a stack of lookups on it belongs in the masking-and-detailing part of your graph.

If you need more than one image's metadata - a whole folder of them - the pack's own Metadata Extractor nodes (single and list) are the siblings for that job. This node is the one-image-at-a-time tool.

Install

ComfyUI Manager → search ComfyUI-mnemic-nodes, or:

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

Restart ComfyUI. Nothing downloads for this node; the metadata parsing is pure Python, though the pack's requirements.txt pulls in piexif, transformers, opencv-python and friends for other nodes at install time.