Blend Images

So you've generated two images and you want a crossfade between them. That's this node. In ComfyUI, ImageBlend ("Blend Images") is the one-node answer to "I like A's composition but B's colors" - it mixes two IMAGE tensors under one slider, no VAE re-encode, no resampling, no models involved. It has been in core since April 2023, which makes it one of the oldest image filters in the whole graph, and it's the kind of quiet workhorse you end up reaching for constantly.

Where it lives in the workflow

This is a pixel-space filter, so it sits after VAE Decode and before Save Image / Preview. Feed it two decoded images and it hands you one. The classic trick: generate the same prompt twice with different seeds, then blend them 50/50 to kill the worst of either seed's weirdness. Or run your normal pass and an upscaled/sharpen pass, then crossfade to get the detail of the second without the overcooked look of a full swap. You can even chain two blends to do "60% A + 25% B + 15% C" three-way mixing, which ComfyUI doesn't give you natively.

How it works

The math is embarrassingly simple, which is why it's fast and why you never see it in "missing node" complaints. image2 gets combined with image1 under the selected blend mode, then the whole thing is lerped toward image1 by blend_factor. Everything is clamped to [0, 1] at the end.

Two behaviors are worth knowing before they surprise you:

• Size mismatch is handled for you. If image2 isn't the same resolution as image1, it's resized to match - bicubic, center-cropped. That's usually what you want, but if your second image is a completely different aspect ratio, the center-crop means you lose the edges. Crop before blending if that matters.
• Channel mismatch is handled too. If one image has an alpha channel and the other doesn't, ComfyUI pads or truncates to match. This got real fixes in April 2026, so on current builds RGBA-vs-RGB blends just work.

The inputs that matter

• image1 / image2 - the two IMAGE tensors. Order matters: the factor blends toward image1.
• blend_factor - 0–1, default 0.5. 0 gives you pure image1, 1 gives you the blend-mode result.
• blend_mode - the one to actually play with: normal (straight crossfade), multiply, screen, overlay, soft_light, and difference. multiply darkens (great for texture stamping), screen brightens, difference is the trickiest - it highlights exactly where the two images differ, which makes it a legit before/after comparison tool.

The single IMAGE output wires straight into any other image node - Save Image, a second blend, or a mask operation.

Getting it

Nothing to install. It ships with ComfyUI core - search "Blend" or "ImageBlend" in the node menu. It's been there since the post-processing nodes landed in 2023.

Common gotchas

• difference mode looks like a glitch if you expect a blend. It's subtraction, so it's for finding differences, not for a pleasing mix. Don't dial it in and wonder what broke.
• Batch frames must line up. If you're blending two video sequences frame-by-frame, the batch dims need to match; a 24-frame clip against a 1-frame image blends that single frame into every output.
• The output is clamped. Overlapping two bright images in screen mode can blow out to pure white - that's the mode doing its job, so lower the factor.