Modelverse Image Packer

This is the least glamorous node in the pack and the one the multi-image workflow quietly depends on. Modelverse Image Packer does exactly one thing: it takes up to five images (or image lists) and stacks them into a single batched IMAGE tensor so the Flux Kontext Pro and Kontext Max nodes can run their multi-image mode. No API key needed, no cloud call, nothing clever - it's pure local tensor plumbing, and that's precisely why you need to know about it.

Why it exists

The Kontext nodes decide between single- and multi-image editing by looking at the shape of the tensor on their images input. One image in → single-image mode. Several stacked → multi-image mode. The packer is how you make "several." You drop up to five IMAGE sources into its inputs, it concatenates them along the batch dimension, and the result feeds directly into the images input of a Kontext Pro or Kontext Max node. That's the whole job - the README's multi-image example workflow is exactly this node feeding the Max editor.

The inputs

• images1 - required. The first image (or list) to pack.
• images2 through images5 - optional. Add more sources up to five total; unused inputs just stay empty.

Output is images - one batched IMAGE tensor containing everything you fed in, in order.

That's it. No model, no parameters, no secrets. If this is the only node of yours that "works," you've done nothing wrong.

Wiring it up

Connect it like this: each source (a Load Image node, a generation node's output, whatever) goes into an imagesN input, and the single images output runs into the Kontext node's images input. The order you attach them is the order they'll be seen - put the primary image first, references after.

Installing

Same install as the whole pack - it's one plugin:

cd ComfyUI/custom_nodes
git clone https://github.com/ucloud/comfyui-plugin

then restart ComfyUI, or use ComfyUI Manager and search "ComfyUl-UCloud". Dependencies are the pack's usual light set (requests, pillow, numpy, openai, GitPython). No API key needed for this node specifically - it never touches the network - but the Kontext node you feed into it definitely does, so you'll want the Modelverse Secret Client wired up anyway.

Common issues

• Multi-image mode doesn't trigger - the most common miss. If your packer only has one source, you get a batch of one, and the Kontext node treats it as single-image mode. Two or more sources to actually get multi-image.
• Images come out in the wrong order - the packer preserves input order; swap the imagesN connections, not the images.
• "At least one image is required" - images1 is mandatory; you can't leave the node empty.
• It's not a compositor - it stacks images into a batch, it doesn't blend or arrange them onto a canvas. For that you want a different node entirely.

For a node with one function, it's easy to dismiss as filler. But the whole "combine these five frames into one edit" capability of this pack flows through it - worth knowing the name when a workflow drags one in and you're left wondering what the gray box does.