MultiPreview

One preview for all your parallel branches

You're running a workflow with three samplers in parallel - different seeds, two checkpoints, one branch testing a LoRA - and you want to watch all three at once. The stock answer is a Preview Image node per branch, which means three floating windows scattered across a graph that's already hard to read. By branch five your screen looks like a used-car lot.

MultiPreview is the fix for that specific annoyance. Drop one in, wire each branch's IMAGE output into image1, image2, image3, and you get a single node with a numbered pin per input. It's a preview/output node, so it has no outputs and sits at the end of a branch like a Preview Image - only it can be the end of a dozen at once.

The clever part is how it stays live. ComfyUI executes depth-first and only refreshes a preview node when it finally runs, which is after every upstream branch has finished. MultiPreview cheats: its frontend extension patches the queue and, at execution time, injects one invisible MultiPreviewInternalReceiver per connected input. Each receiver saves its branch's image to ComfyUI's temp directory the moment that branch completes and pushes it to the browser over a custom event. So when branch 2 lands, you see branch 2 - you don't wait for the other five. The parent detects the injected receivers and skips its own redundant save; drive ComfyUI as a pure API with no browser, and it falls back to a normal preview save and still works.

Inputs are the whole story, and there are 32 of them:

• image1 … image32 - optional IMAGE inputs. Wire whatever branches you care about and leave the rest dangling; a new empty input appears automatically as you connect more.

On the node you get numbered pin buttons to switch what's displayed, plus an auto_latest toggle that follows the newest finished input instead. Batch results use the standard ComfyUI image controls, with one genuinely nice touch: each pin remembers its own batch position, so flipping between pins doesn't lose your place in a 40-image batch. And preview state restores when you switch workflow tabs or views during the same browser session - leave and come back, the thing you were looking at is still there.

Install is as easy as this pack gets, because there's nothing to install:

cd ComfyUI/custom_nodes
git clone https://github.com/palealloy2999-prog/comfyUI-multi-preview

Restart ComfyUI, then hard-refresh the browser (Ctrl+Shift+R) so the JS extension actually loads. Or search "comfyUI-multi-preview" in ComfyUI Manager and click install. Zero Python dependencies, no model files - the whole pack is a thin frontend layer over ComfyUI's built-in PreviewImage, published by palealloy2999-prog through the official Comfy registry under an MIT license. This is not a pack to fear on the dependency-hell front.

Where people get burned: it's a preview node, so nothing you see here is saved - previews live in temp/ and UI state lives in the browser session. If you like a result, put a Save Image behind that branch before you close the tab. Don't panic if the node turns red with "Required input is missing: images" - that's ComfyUI's normal reaction to running a preview node with nothing connected. And if the pin buttons never appear after an update, the culprit is almost always a stale browser tab; the hard refresh is not optional.

Honest take: this is a niche quality-of-life node, not a workflow essential. But if you live in multi-branch workflows - LoRA comparisons, seed sweeps, model A/B tests - it turns a wall of floating previews into one tidy place to look. That's worth the two minutes of install.