Load Image

ComfyUI's built-in Load Image is fine. It loads an image. But if you've ever wanted to know an image's dimensions inside the graph, or extract the metadata that's embedded in a PNG you downloaded from Civitai, the stock node makes you add extra nodes for each thing. Load Image from Sage Utils is the version that hands you everything at once: the image, its alpha as a mask, its width, its height, and its embedded metadata - five outputs off a single node.

It's one of those "why isn't this in core" quality-of-life nodes, and it's a good entry point to the pack's whole philosophy: fewer nodes doing the same work, more information flowing through the graph.

How it works

Like the core loader, it lists the files in your ComfyUI input directory in a dropdown (image_name), and it supports dragging in or uploading a new file. Run it and you get:

• image - the loaded IMAGE.
• mask - the alpha channel as a mask, when the file has one. Transparent PNGs come through as masks you can use for compositing or inpainting without a separate mask-extraction step.
• width / height - the dimensions, as plain INTs. Wire these into an Empty Latent or a resize node and your workflow can adapt to whatever image you loaded instead of assuming a size.
• metadata - the embedded metadata string. If the image came from A1111 or Civitai, this carries the original prompt and settings - genuinely handy for reverse-engineering or for passing provenance into Sage's metadata nodes.

Two implementation details worth knowing: the file list is cached for a few seconds so the node doesn't rescan the input directory on every call (which would slow down big graphs), and it validates that the file actually exists before running, returning a clean "invalid image file" message if you've pointed it at something missing.

Wiring it up

The classic setup: load a reference image, feed width/height into an Empty Latent or into a resolution helper like the pack's Guess Resolution By Ratio to normalize for generation, and pipe image into an img2img or ControlNet branch. If the file has a workflow embedded in it, metadata is where you'd pull it from for display or re-embedding.

Installing

Part of Sage Utils:

cd ComfyUI/custom_nodes
git clone https://github.com/arcum42/ComfyUI_SageUtils.git
cd ComfyUI_SageUtils && pip install -r requirements.txt

Restart ComfyUI, or install "Sage Utils" via ComfyUI Manager. No models needed - it's a pure utility.

Where people get tripped up

The mask output is the alpha channel, not a separately-authored mask file. If your image has no transparency, that output is effectively empty - don't go looking for a mask that was never there. Also, the dropdown lists files in the input directory at graph load, so if you drop a new file into input while ComfyUI is running, it may not appear until you reload the node or refresh. And metadata is a raw string - for A1111/Civitai images it's usually a parameters JSON or plaintext block, which is readable but not parsed. That's what the pack's dedicated metadata nodes are for; this output just gets you the raw text.