Image Compare

The still-image sibling of this pack's KY_VideoCompare, and it works on exactly the same idea: drop two images in, get a draggable before/after slider right on the canvas instead of alt-tabbing between two separate previews. Zero outputs, by design - this is a look-at-it node, not a step in a pipeline.

What it accepts

Every input on this node is optional, and they come in two independent pairs. image_a_url_or_filepath/image_b_url_or_filepath take a plain string - a URL or a local path, typed or pasted directly. image_a/image_b take a proper IMAGE-typed socket wired from an upstream node. You can mix them freely: paste one image's URL directly while wiring the other from a generation node's output, whatever's convenient for the comparison you're actually trying to make.

Because nothing is required, you could technically run this node with everything blank, though obviously that gets you nothing useful to look at - in practice you'll want at least one image on each side of the comparison for it to do its job.

Where you'd use it

Anywhere you'd otherwise be squinting between two saved files: comparing a base image against an inpainted or upscaled version, checking two different seeds or samplers against each other, or verifying a KY_LoadImageFrom/KY_ReadImage result actually matches what you expected against a reference. It's a QA tool for a human looking at the ComfyUI canvas, not something you'd wire into an automated pipeline - there's nowhere for a "winner" to flow out to, since it has no outputs at all.

Installing it

ComfyUI Manager: search ComfyUI-KYNode, install, restart. Or:

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

Restart ComfyUI. No external dependencies for the comparison UI itself. If you're feeding it a URL rather than a local path or a wired IMAGE, that does need network access from wherever ComfyUI's process runs - same class of requirement as this pack's URL-based image loaders.

Common issues

If you meant to compare two images but only see one, double-check you actually filled in both sides of one of the two input pairs - since the string and IMAGE pairs exist independently, it's easy to wire image_a from a node but only type a path into image_b_url_or_filepath, forgetting they don't need to match modes, but each side still needs something.

Because this is a preview-only node with no outputs, don't expect to chain anything after it - if you're trying to programmatically pick a "better" image and route it downstream, this isn't the right tool; it's built for eyeballing, not scoring.

If a URL-sourced image doesn't load, treat it the same as any broken image link - check that the URL is actually reachable from wherever ComfyUI is running (a URL that opens fine in your browser isn't proof the server process can reach it, especially behind a firewall, VPN, or containerized deployment).