ImageBlend(Badman)

ComfyUI's core image-blend node covers the basics, but if you've ever reached for screen or soft_light in Photoshop and gone looking for the equivalent in a graph, this is that node. It's a straightforward two-image blend with a strength slider and a real set of standard compositing modes, not just a normal-mode crossfade.

What it's for

Blend modes are the bread and butter of compositing: laying a texture or effect layer over a base image without it looking like a flat overlay. multiply darkens by combining the two images' shadows - good for adding grime, vignettes, or shadow passes. screen does the opposite, brightening by combining highlights - good for light leaks, glows, or lens-flare-style overlays. overlay and soft_light both punch up contrast by combining multiply and screen depending on the base image's own brightness, with soft_light being the gentler of the two - useful for subtle grading passes that shouldn't blow out highlights or crush shadows. difference is the odd one out: it doesn't composite so much as measure, turning identical regions black and different regions bright - genuinely handy as a debugging tool to diff two renders and see exactly where they diverge, not just an artistic effect. add is the blunt instrument, summing pixel values for a hot, overexposed look. normal is the plain crossfade you'd expect as a baseline.

Typical use: generate a base image, generate or load a second image (a texture, an effect pass, a second seed of the same prompt), and blend them at a controlled strength rather than fully committing to either.

Inputs and outputs

Four required fields:

• image1, image2 - the two IMAGE inputs being combined. image1 is the base, image2 is the layer being blended on top, though which reads as "on top" depends on the mode you pick.
• blend_factor (float, 0–1, default 0.5, step 0.01) - how strongly image2 is blended in. 0 gives you back image1 unchanged; 1 applies the blend mode at full strength.
• blend_mode - an enum with seven options: normal, multiply, screen, overlay, soft_light, difference, add.

One output: the blended IMAGE.

Installing it

ComfyUI Manager: search ComfyUI-BadmanNodes, install, restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/MariusKM/ComfyUI-BadmanNodes

No dependencies or model downloads called out in the README - this is pixel math on tensors you already have loaded, so it's a light, fast install with nothing extra to fetch.

Common issues

The two images need matching dimensions (and ideally matching batch size) going in - mismatched resolutions between image1 and image2 are the single most common reason a blend node throws a shape error, so if you're blending a generated image against a loaded texture, resize one to match the other first. Beyond that, the modes that key off relative brightness - overlay and soft_light especially - will look completely different depending on which image you put in image1 versus image2, since they're not symmetric operations; if a blend looks inverted from what you expected, try swapping the two inputs before assuming the mode is broken. And difference mode is worth remembering exists even outside compositing work: pairing it with blend_factor at 1 and two otherwise-identical renders is a fast, visual way to spot exactly what changed between two generations without eyeballing a side-by-side.