Blend Images

IsekaiBlend is the compositing node from the isekai-comfy-node grab-bag, and honestly it's the one you'll reach for first out of that pack. It takes two images, slaps them together with one of eight blend modes, and hands you a single combined image. In a ComfyUI graph you'd normally have to build this yourself with a bunch of math nodes, or pull in a heavyweight pack just to get a multiply. This one is a single node with a dropdown. Sometimes simple wins.

What it does and why you'd bother

The classic uses: composite a logo or watermark onto an image (Multiply or Screen depending on whether your logo is dark or light), crossfade between two generations, or overlay a color grade. The "Normal" mode with opacity below 1 is just a plain transparency blend, which makes this a handy poor-man's image mixer - feed it the same image twice at opacity 0.5 and you've got a cheap fade, though nothing beats real interpolation for that.

Mechanically it's PIL all the way. Both IMAGE tensors get converted to Pillow images on the CPU, image_b is resized to match image_a if their dimensions disagree (so you don't get a size-mismatch error - it just quietly squishes one), the blend mode runs through PIL's ImageChops, and then opacity cross-fades the result back toward image_a via Image.blend. No GPU involved, no model involved, nothing to download. It's the kind of node that just works.

The inputs that matter

• image_a and image_b - your two images. Order matters for Subtract and Difference.
• blend_mode - Normal, Multiply, Screen, Add, Subtract, Difference, Lighten, Darken. Multiply darkens, Screen lightens, Difference gives you that inverted "how different are these two images" look.
• opacity - 0 to 1, default 1. At 1 the blend mode is applied at full strength; below that it fades toward image_a.

Output is a single image tensor, which you can wire straight into a Save Image node or on into more color work.

Installing it

This comes with the whole pack, so there's no separate install:

Via ComfyUI Manager: open Manager → Install Custom Nodes → search "isekai" → install isekai-comfy-node, then restart ComfyUI.

Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/isekai-sh/isekai-comfy-node
cd isekai-comfy-node
pip install -r requirements.txt

Then restart. You'll find it under the Isekai → Image/Blend menu. Dependencies are just Pillow and requests (torch and numpy ship with ComfyUI already) - no model downloads, nothing heavy.

Where people get burned

One real gotcha: the underlying tensor-to-PIL helper only processes the first frame of a batch. If you feed it a batched image and expect every frame to be blended, only frame zero comes back. Fine for single images, wrong for batch workflows. Also note everything runs in RGB, so any alpha channel gets dropped - if you're compositing with transparency you'll want to do that elsewhere or re-add the mask after.

A small nitpick: if the pack's console spam bothers you, every node prints [Isekai] ... lines to the terminal. Harmless, but noisy if you're used to clean logs.