Image Concatenate (UTK)

The core ComfyUI Image Concatenate node is famously finicky - it wants the two images to already match on the stitching axis, and if they don't, you're resizing by hand. Image Concatenate (UTK) from ComfyUI-UniversalToolkit removes that chore: it resizes the images to a common dimension for you, so two mismatched generations just go together. For before/after shots, batch comparison strips, or assembling reference sheets, this is the version I'd put in the graph.

How it works

You feed image1 and image2. If match_image_size is on (it is, by default), the node resizes both so they line up on the join - for a horizontal right stitch, it matches heights; for down, it matches widths - using bilinear. Then it concatenates them onto a canvas, adding a gap in pixels if you set one, and capping the result at max_size (default 4096) by downscaling if the combined canvas would blow past it.

The smart touch is direction = "auto". Rather than forcing you to guess whether your images are taller or wider, it looks at the two dimensions and picks the axis that makes the canvas most square-ish. Two 1024×1024 squares go side by side; two tall portrait crops stack vertically. You can override with explicit right/down/left/up if auto doesn't match your intent. background_color (black/white/gray/transparent) fills the gap and any leftover canvas.

Inputs and outputs that matter

• image1 / image2 - the two frames. Both required.
• direction - auto (default) or right/down/left/up.
• match_image_size - resize to a common join size. Leave on.
• gap (0–512) and max_size (64–8192) - spacing and output cap.

Single output: IMAGE, the stitched frame. It's a normal (non-output) node, so route it into a preview or save node yourself.

Install and gotchas

Install via ComfyUI Manager (search "ComfyUI-UniversalToolkit") or:

cd ComfyUI/custom_nodes
git clone https://github.com/whmc76/ComfyUI-UniversalToolkit
pip install -r requirements.txt

Restart ComfyUI. No models, no keys - Pillow/numpy/torch.

The honest caveat: "match sizes" here means stretch to match, not aspect-crop. If you stitch a 4:3 against a 16:9 horizontally, the 4:3 gets stretched to the 16:9's height, which can visibly fatten faces. For outputs of the same model at the same resolution - the normal case - you'll never notice. For mismatched material, normalize aspect ratios first. And one more: the output dims are capped by max_size, so if your two images are already 4096 each, the combined 8192 canvas gets scaled down to fit. That's a feature for sanity, but it means the stitched result won't be true 1:1 pixels for very large inputs.