Batch Image Blend

Where its sibling Multiple Image Blend takes up to four separately-named image inputs, this one takes a single images input - a whole batch - and flattens the entire thing into one output image, weighted by one opacity value. No named slots, no per-image control: whatever's in the batch gets combined.

What it's for

Anywhere your graph has already produced a batch of images and you want them reduced to one, this is the node for it. The obvious case is averaging out seed variance: run the same prompt through a KSampler with batch_size set above 1, feed all of those outputs into this node, and get back one blended composite that smooths over the per-seed randomness - useful for previewing "the general shape of what this prompt tends to produce" without picking a single lucky seed. It's equally usable on a batch that arrived some other way - frames pulled from a short animation, several passes of the same region - anywhere "combine this batch into one image" is the actual ask.

This is distinct from Multiple Image Blend, which wants four individually wired image inputs rather than one batch tensor. Use this one when you already have a batch; use the other when you're composing a handful of separately-sourced images by hand.

Inputs and outputs that matter

• images (required) - a batch of images, IMAGE type. However many images are in the batch get combined into one.
• opacity (required, 0–1, default 1, step 0.01) - controls how strongly the batch's combined content comes through in the result.

The output is a single image - the flattened result, ready for a Save Image node or further processing.

Installing it

ComfyUI Manager: search ComfyUI-ReplenishNodes, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/hben35096/ComfyUI-ReplenishNodes

No model downloads or extra dependencies documented - it's straightforward image-processing logic.

Where people get tripped up

Mismatched sizes inside the batch. As with its sibling, there's no resize or alignment control here. A batch that isn't already uniform in resolution (which shouldn't happen if it came straight out of a single KSampler run, but can if you've concatenated images from different sources into one batch) is likely to error rather than blend gracefully. Keep the batch dimensionally consistent before it reaches this node.

Expecting weighted averaging by position. The opacity control is a single shared value for the whole operation, not a per-frame weight - you can't make image 1 in the batch count more than image 5. If you need that kind of control, you'll need to split the batch and blend pieces individually with different opacity settings.

A batch of one. With a single-image batch, there's nothing to blend against, so the output is effectively that same image passed through. That's expected behavior, not a malfunction.