Image Load with Metadata (WLSH)

The pack's author (Reddit handle wallish, GitHub wallish77) built this whole node pack after moving from Automatic1111 to ComfyUI around SDXL's release, and this node is the one he specifically called out as a founding motivator: he wanted metadata read back out of images the way A1111 did it, instead of ComfyUI's all-or-nothing move of dropping a PNG onto the canvas and replacing your entire workflow. This node does the narrower, more useful thing - it reads an image's embedded prompt/sampler data and hands you the individual pieces as outputs you can wire wherever you actually want them.

How it works. It reads from ComfyUI's own input folder - same as the stock Load Image node, you pick a file from a dropdown rather than piping an IMAGE in from elsewhere - but instead of just handing back pixels, it also parses whatever generation metadata is embedded in the file. That works on images saved by Automatic1111 (which embeds prompt/seed/steps/etc. as PNG text chunks) and on images saved by this pack's own save nodes, since they write in a compatible format. Images that never had that metadata to begin with - a screenshot, something re-saved through an editor that strips PNG text chunks, or a file downloaded from a site that scrubs it - will come back with those fields empty. That's not a bug in the node; there's simply nothing there to read.

Inputs: image, the file picker dropdown, and verbose (true/false) - flip that on and it'll print what it found (or didn't) to the console, which is the fastest way to debug why a field is coming back blank. Outputs: image (the pixels, just like a normal loader), plus positive, negative, seed, steps, cfg, width, and height pulled straight from the file's metadata - wire those into a KSampler, a prompt box, or this pack's save nodes to reconstruct or continue from a previous generation without retyping anything.

Installing it

ComfyUI Manager: search "wlsh_nodes" or "WLSH Nodes", install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/wallish77/wlsh_nodes, then restart. Nothing extra to download - this node's whole job is text parsing, not inference.

Common issues

By far the most common complaint is empty outputs, and it's almost always the metadata simply isn't there - check verbose output in the console before assuming the node is broken. The README also notes an old ComfyUI-version requirement for the image preview to show up (a specific commit from ComfyUI's early history); given how far core has moved since, that's essentially a non-issue on any current install, but it's worth knowing it exists as a historical footnote if you're ever running an ancient ComfyUI build. One more thing worth knowing going in: this node loads its own file and reads its own metadata - it doesn't take an IMAGE input from elsewhere in your graph, so you can't chain it after another loader or a generation step and expect it to retroactively find metadata that was never embedded in the first place.