Image Tensor List

If you've ever gotten the "images must be the same size to batch" error, this tiny node is the workaround. ComfyUI has two ways to carry multiple images: a batch (one tensor with all the images stacked, which forces them to share dimensions) and a list (separate items that can each be whatever size they want). Image Tensor List makes the second kind. It bundles images into a list so you can pass differently-sized pictures downstream without the batch dimensions having to match.

It's part of KJNodes, kijai's utility pack - pure plumbing, the kind of node you only learn exists the day you hit the error it solves.

How it works

A batched tensor is efficient but rigid: everything in it has to be the same height and width. A list drops that constraint by keeping the images as distinct entries. This node takes image inputs and returns them as a list, so the next node - as long as it's written to accept a list - can iterate over images of mixed sizes instead of choking on the mismatch.

The inputs and output that matter

It takes image1 and image2 and returns a single IMAGE output that is actually a list of the two. Need more than two? Chain the node - feed the output into the image1 of another Image Tensor List and add a third, and so on - to build up a longer list. The important detail is on the receiving end: the downstream node has to be a list-aware node (one that processes inputs item by item). Wire this into something expecting a plain batched image and it won't behave the way you want.

Installing it

ComfyUI Manager → search KJNodes for ComfyUI → install → restart. Or cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-KJNodes, then pip install -r ComfyUI-KJNodes/requirements.txt, and restart. No models, no heavy dependencies - it's a data-shape node.

Where people get burned

• Batch vs list confusion. This is the whole point of the node and also the thing that trips people up. A list is not a batch; some nodes want one, some want the other. If the "same size" error is your problem, a list is the fix; if a node then complains it got a list, that node wanted a batch and you've over-applied this.
• Downstream has to be list-aware. The list only helps if what comes next iterates over items. Many nodes do, but not all - check that the consumer actually handles a list before blaming the node.
• Order matters. Chaining these builds the list in the order you wire them; if sequence matters downstream, mind which image goes into image1 vs image2.