ImageBlend2

ImageBlend2 is the stock ComfyUI Blend node with five extra blend modes bolted on. If you already know Blend, you know this: two images in, a blend_factor slider, and a mode dropdown - but where the built-in node stops at normal, multiply, screen, overlay, soft_light, and difference, this one adds compare_light, compare_dark, compare_color_light, compare_color_dark, and abs_diff. Output is a single blended IMAGE.

Those extras are the reason to reach for this over the built-in node. The two most useful:

• abs_diff - the absolute difference between the two images, pixel by pixel. Feed it two nearly-identical renders and the output lights up exactly where they changed. It's the fastest way to see what a new sampler, seed, or model actually altered. Think of it as free A/B diffing.
• compare_light / compare_dark - per-pixel pick-the-brighter / pick-the-darker image (what Photoshop calls lighten/darken). Good for compositing or recovering detail that one pass lost.

The two compare_color_* variants do the same pick-the-extreme, but decide using per-pixel color luminance rather than raw channel values, so they're a bit more perceptually sensible when color matters. The blend_factor still applies the same way it does in Blend - at 0 you get image1, at 1 you get image2, in between is the mix - which matters because abs_diff and the compare modes are about which image shows through, and the factor tunes how much.

Mechanically it's a subclass of ComfyUI's Blend: same input structure, same blend_mode enum, with the new modes appended. All five extras are implemented as simple torch tensor ops (torch.where comparisons and torch.abs), so they're cheap and fully GPU-side.

Install via ComfyUI Manager (search "ComfyUI-nodes-hnmr") or:

cd ComfyUI/custom_nodes
git clone https://github.com/CYBERLOOM-INC/ComfyUI-nodes-hnmr

Restart ComfyUI; no downloads, no dependencies beyond ComfyUI's own.

Realistic use: abs_diff in a quality-control loop (blend a batch of variations against the baseline and eyeball what changed), or compare_light for merging a detail pass back into a base render. The one trap is expecting Photoshop semantics exactly - these are straightforward tensor comparisons, so color grading differences between modes are subtle. Test on your actual images before trusting a mode. And as with the rest of this pack, install the CYBERLOOM-INC fork - the original repo stopped getting fixes and broke on modern ComfyUI.