Kimara.ai Batch Images

Sometimes the hardest part of a ComfyUI workflow is a type mismatch: one node hands you a Python list of image tensors, and the next node only accepts a single batched IMAGE. This is that bridge. Kimara.ai Batch Images takes a list of image tensors, stacks them along the batch dimension with a single torch.cat, and hands back one IMAGE. No sliders, no options, nothing to misconfigure. It's the pack's plumbing: the sibling node, Kimara.ai Advanced Watermarker, deliberately outputs a list, and this is the natural thing to plug it into.

How it works

The input port is marked INPUT_IS_LIST, which tells ComfyUI to hand the node the whole list in one call instead of running it once per item. Then it's just concatenation along dimension 0 - image 0, image 1, image 2 become one tensor of batch size N. No resizing, no blending, no per-frame work. It's the most honest node you'll meet.

The one thing that bites

Read the source and you'll find the return only happens when there's more than one image in the list. Feed exactly one image and the node returns nothing, which ComfyUI will generally choke on. Practical rule: this node only does anything for two or more items. If you're holding a single-image list, skip it and pull the tensor off directly.

Second, because it's a plain torch.cat, every image in the list must share the same height, width, and channel count. Mixed resolutions will raise. That's the normal case when the Watermarker built the list, so in practice you're fine.

Install

Same story as its pack-mate, and it's the easy kind of install: requirements.txt is just Pillow and numpy, both already present in a stock ComfyUI, and there are no model downloads. Grab it via ComfyUI Manager (search "Kimara" or "Advanced Watermarking"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/kimara-ai/ComfyUI-Kimara-AI-Advanced-Watermarks

then restart ComfyUI. It's AGPL-3.0 licensed.

When to reach for it

• Right after KimaraAIWatermarker, its intended partner - watermark a batch, then this node re-batches the list for whatever comes next.
• Any time a custom node hands you a list of images - some samplers and preview utilities do - and your next node wants a single batch tensor.
• Before a Save/Preview node that only accepts a plain IMAGE.

That's the whole thing. It's a two-line utility that exists to fix an impedance mismatch, and it does exactly that one job.