AD Image Concatenation Advanced

"AD" here is Addoor, not Adobe - this is a cleaned-up fork of the padding/concat node from Eagle-CN's Addoor pack, and honestly it's the version worth using. Plain image concatenation sounds like the kind of thing ComfyUI should already do natively, but core ComfyUI makes you reach for one-off utility nodes for even a simple side-by-side. This one takes up to five images, jams them together in a row or column, and hands you back a single IMAGE tensor. That's it. No model files, no API, no hidden dependencies - the whole thing is torch, torchvision and PIL, which ComfyUI already ships.

Where it earns its keep: before/after comparison shots for an upscale or face-fix workflow, a quick 2-3 panel grid you're about to post, or - the reason BigStationW builds these - feeding multiple input images into a multi-image edit model like Flux Klein. There's a reason workflow-sharers keep a node like this in their graph.

The mechanism is refreshingly transparent. Each input tensor becomes a PIL image, the ones you want to match get resized, everything is center-aligned onto a fresh canvas, and the result converts back to a tensor. The output is always a list (the schema marks it as such), so wire it into a Preview or Save node and ComfyUI iterates for you.

The inputs that matter are only a few. image1 is required and feeds through optional image2–image5; leave the ones you don't need disconnected. direction flips between horizontal and vertical. match_size (on by default) resizes the earlier images to match the later one's height - so if you want a specific image to set the size, wire it in last. When it's off, mismatched images just center-align inside a taller/wider canvas. method picks the resample filter; lanczos is a fine default, nearest only if you're doing pixel-art. And output_all_concatenations is the genuinely clever bit: it returns every progressive step as a list - image1, then image1+2, then image1+2+3 - so you can preview the build-up or feed it to a Save node to keep each stage.

Where people get burned: it only uses the first frame of each input batch. Feed it a 4-image batch and you get one image back, silently. And match_size stretches the earlier image to fit the last one's dimension - put the master image anywhere but last and you'll wonder why things look squished.

Install is the standard two-liner:

cd ComfyUI/custom_nodes
git clone https://github.com/BigStationW/Comfyui-AD-Image-Concatenation-Advanced

Restart ComfyUI, or just search the pack title in ComfyUI Manager and click install. No models to download, nothing else to satisfy - it's one file. If a script errors after install, it's usually a batch being fed where a single image is expected. Which, as noted, is this node's one real trap.