Unpack Image Collection

Say you generate a character sheet, or four poses, or forty variations, each from its own sampler - and now you want to run them all through one batch operation (upscale, video-feed, color-correct, whatever). The problem is they're scattered across different nodes with different dimensions and different channel counts. Unpack Image Collection is the gather point: it collects every image you've tagged with a shared name and stacks them into a single uniform batch.

This is the receiver half of a two-node pair in Mickmumpitz-Nodes. The sender is Image Collect, which sits on an image output and just tags it (it's a pure pass-through - the image goes through unchanged). Give several Image Collect nodes the same name, then this node selects that name and every matching image arrives as one images batch, resized to a common size.

How it works

The mechanism is the pack's favorite trick: a queue-time prompt handler. When you submit, the handler finds every Image Collect node with the matching name and injects its image into one of the 128 hidden input_N slots this node declares. No wiring needed on your part - you just set source to the shared name and connect the output.

Because images land from different generators, they need unifying. That's what the two enum settings do:

• target_size - whose dimensions win: first (the first image collected), largest, or smallest.
• fit - how each image is squeezed into that size: stretch (ignores aspect ratio), pad (fits inside, black bars), or crop (covers, center-cropped).

Channels are padded too - if one input is RGB and another is RGBA, everything becomes RGBA so the stack is valid.

Inputs and outputs

The only things you'll actually touch:

• source (STRING) - the shared name, must match the Image Collect tags you want to gather. This is the one to get right; a typo and you collect nothing.
• target_size and fit - set once for your workflow; first + stretch is the fastest default if all your outputs are already the same size.

Output is a single images IMAGE tensor (a batch), which wires straight into whatever batch-processing node you were pointing at one image before.

Where it fits

This shines in the iterative/consistent-character workflows this pack is known for - Mickmumpitz is a YouTuber whose character-sheet and video setups generate lots of parallel branches that need recombining. Collect-then-unpack is also handy for feeding a video model a stack of context frames you built in several separate runs.

Installing

One pack, one install: ComfyUI Manager → search "Mickmumpitz" → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/mickmumpitz/ComfyUI-Mickmumpitz-Nodes

Just numpy, Pillow, opencv-python under the hood. There's a companion JS file (collect.js) that the pack loads automatically, so the tag widgets work out of the box.

Gotchas

If nothing shows up, you almost certainly have a name mismatch between the Collect and Unpack nodes, or the Collect nodes were created after the workflow was last queued - re-queue and it should appear. If resolutions differ wildly, pad and crop behave very differently from stretch, so pick deliberately. And there's a ceiling: up to 128 tagged images can be gathered, which is generous, but a runaway loop of Collects can hit it.