🦊 RS Image Compare

You know the drill. Two seeds, two models, two denoise values, and you're flipping between previews trying to remember which one had the cleaner eyes. 🦊 RS Image Compare (class RSComparer) is the antidote: drop two images in, drag a divider, and zoom to 10x when you finally find a difference worth inspecting. It's an output node - it doesn't transform anything, it just shows you both images in one panel with a slider between them. Think of it as rgthree's image comparer, but with heavier zoom and pan controls.

How it works

This is a pure display node, so the mechanism is simple: you wire up to two images, and the frontend renders them overlaid with a vertical divider. Drag the divider left and you see more of image_2; drag it right and you see more of image_1. Pan and zoom are computed relative to the image center, and pan values are percentages of the zoomed dimensions - so at 2x, a pan of 50 moves you half an image-width, not half a pixel. New images auto-reset all the controls, which is the behavior you actually want when you're A/B-ing a dozen variants in a row.

Inputs that matter

• image_1, image_2 - the two IMAGE tensors you're comparing. Both are optional, so you can hook up one and it still renders.
• zoom - 0.1x to 10x, step 0.1. This is the whole point: get to 10x and you can tell if that "detail" is real or a sampling artifact.
• pan_x, pan_y - integer pan offsets in percent, −5000 to 5000. Pair them with zoom to wander around the image.

The single output, images, passes the input through as an IMAGE tensor - handy if you want to preview mid-workflow and still keep the data flowing. It's marked as an output node, so it works as a terminal preview like the native PreviewImage.

How to install

It ships in ComfyUI_RaykoStudio, so you get it with the whole pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Raykosan/ComfyUI_RaykoStudio.git
# restart ComfyUI

Or search "ComfyUI_RaykoStudio" in ComfyUI Manager and install from there. The pack lists torch, opencv-python, numpy, Pillow, freetype-py, pycairo, and scipy in its requirements - you almost certainly have most of them already, and the compare node itself has no special dependencies.

Common issues

• Nothing shows in the panel. You've wired image_1 and image_2 but haven't run the queue yet, or one of them is disconnected - the node only updates when it executes. Re-queue and it'll refresh.
• Pan feels enormous at high zoom. Remember pan values are relative to the zoomed image, not the original. At 10x you'll want tiny values.

The one honest caveat: this node is a nicety, not a necessity. If you only ever compare a couple of generations, a browser tab with both previews does most of the job. Where it earns its keep is in a big batch workflow where you're rapidly flipping between candidate outputs - that's exactly what its auto-reset-on-new-image behavior is built for.