Batch to Multi Image Output(TJ)

Batch to Multi Image Output (TJ) is the un-batcher: it takes one IMAGE batch and splits it into individual IMAGE_1 through IMAGE_64 outputs so each image can go down its own branch. The use case is anything where you want to treat images differently after a batch step - upscale only the good ones, run a face-fix on one, save the rest as-is, compare them side by side. A normal batch forces every downstream node to process every image; this node hands you per-image lanes.

How it works

You feed it a batch via the images input and set out_count to however many output lanes you want (1–64). On execution it slices the batch: image i of the batch goes to IMAGE_i+1. Three details matter:

• Outputs beyond the batch size are not errors - each unused lane gets a black placeholder tensor matching the batch's resolution. If your batch has 4 images and out_count is 6, IMAGE_5 and IMAGE_6 come out black.
• It always declares all 64 outputs in the API, so anything you leave unconnected is simply dropped - you can set out_count to 2 and ignore the other 62.
• get_name is the wireless receive slot: pick a TJ Set provider and the batch arrives without a wire. auto_set does the reverse - publish each output as a wireless provider.

So the honest mental model: this is a batch de-multiplexer with training wheels. If you know the batch size at workflow-edit time, you can hardcode out_count; if it varies, keep in mind that "missing" lanes show up as black images, not None - and anything that takes an IMAGE will happily process a black frame. (The pack ships a derived node, Batch to MinimaxH3, that returns None instead of black for overflow, specifically because MiniMax H3 reference slots treat None as "disconnected".)

Inputs and outputs that matter

• images - the batch to split.
• out_count - how many lanes to activate. Set it to your expected batch size.
• get_name / auto_set - wireless receive and publish.

Outputs: IMAGE_1 … IMAGE_64, one single-image tensor per lane. Wire each into whatever branch you want, then into a preview or save node.

Install

ComfyUI Manager → Install Custom Nodes → search TJ_NODE, or:

cd ComfyUI/custom_nodes
git clone https://github.com/designloves2/ComfyUI-TJ_NODE

Restart ComfyUI. Nothing extra to download for this one. Category: ✨ TJ_Node/Image.

The trap to dodge

The black-placeholder behavior is the thing that bites people. If you're splitting a variable-size batch and wiring all lanes, you'll get black frames in your outputs whenever the batch is shorter than out_count, and nodes downstream will happily save or compare those black frames. Either set out_count to match your known batch size, or gate the branches so a black placeholder never reaches a save node. That's not a bug - it's the design - but you need to know it's there.