Image Compare

You've got two near-identical images - same seed, one setting nudged - and you need to see the difference. Flicking between two tabs works until the change is subtle, and then your eyes give up. This node is the fix: feed it two images, hit run, and drag a slider across a side-by-side. Image A lives left of the line, image B right, and the diff jumps out the moment you drag. There's no API, no key, no model to download - it's just a clever A/B display, and that's the whole point of it.

It's one of the rare ComfyUI nodes that's pure presentation. Two IMAGE inputs, image_a and image_b, and nothing out. It's an output node: the comparison renders in its own body and that's the entire job. You don't wire anything out of it, you just look at it.

How it works

The Python side takes the first frame of each input tensor, converts it to a PIL image, base64-encodes it as a PNG, and hands both to the frontend. The JavaScript side then draws image B across the full preview and paints image A clipped to the left of the slider line. You can grab the cyan handle and drag, or just click anywhere in the preview to jump the slider there. It even labels the corners A and B and prints each image's dimensions - genuinely useful when the two inputs don't match in resolution, which brings up the one real gotcha:

• Only the first frame of each input is compared (image_a[0]). Feed it a batch and you'll see frame one, not a montage.
• Each image is scaled "contain" style and centered on its own. If the two images differ in aspect ratio they won't line up pixel-perfect, so for a true diff you want both at the same dimensions - which is exactly the "same seed, one change" use case it's built for.

Install

Install through ComfyUI Manager (search "image compare"), or the old reliable way:

cd ComfyUI/custom_nodes
git clone https://github.com/Sean-Bradley/ComfyUI-Image-Compare.git

Then restart ComfyUI. The node shows up under the SBCODE category, or double-click the workspace and search image compare. There are no extra dependencies to juggle - the pack adds nothing beyond PIL and numpy, which ComfyUI already ships.

Common issues

It runs but the preview stays blank. The known culprit in late 2025 through 2026 is ComfyUI's Nodes 2.0 frontend rewrite, which broke a whole family of custom-JS nodes - this one, rgthree's comparer, and others. The fix people actually hit on: turn Nodes 2.0 off and restart. It's the same workaround the community settled on for rgthree's pack, so if the node worked last month and broke after an update, suspect that before anything else.

You ran it and "nothing happened." Remember it's an output node - the result isn't a tensor going anywhere, it's a picture drawn in the node body. Queue it, let execution finish, then look at the node itself.

The honest take: this is a lean, single-purpose alternative to rgthree's famous Image Comparer. If you already run rgthree-comfy (and a lot of people do, just for its other utilities), you may not need this node at all. But when you want one small tool that does exactly one thing, this is it - and it comes from Sean Bradley (SBCode), a long-running ComfyUI educator whose tutorials people still recommend when a beginner asks where to learn.