AK Index Multiple

Batch nodes give you lists; samplers want single items. AKIndexMultiple is the bridge. Feed it any list - images, masks, latents, prompts, whatever - and it hands you the items you asked for as individual outputs, each on its own wire. It's the de-spaghetti tool for workflows where you batch-load a folder of reference images and then want to do something different with each one.

The idea is the same as the sibling AKCLIPEncodeMultiple node, except that one CLIP-encodes its slice for you. If your list is prompt strings you want turned into conditioning, use that; if it's anything else - or you want the raw values - use this.

How it works

At its heart it's a range extraction. You give it:

• input_any - the list. Images, latents, masks, text, numbers, whatever; the node is typed *.
• starting_index - where in the list to start slicing (0 = first item).
• length - how many items to pull out, up to 50.
• if_none (optional) - a fallback value for slots that come up empty.

Then it fills its outputs item_0 through item_49. Here's the thing that trips people up: the node always exposes 50 outputs, but only length of them carry real data. You wire up as many as you asked for and ignore the rest. Set starting_index=0, length=4 and you'll use item_0–item_3; the remaining 46 outputs sit there unconnected and unused.

Anything outside the slice - either past the end of your list or beyond length - gets the if_none fallback if you provided one, or None if you didn't. That's deliberate: it means a short list won't make the node error out, it just yields empty slots you can catch downstream.

One implementation detail worth knowing: the node runs in list-input mode (INPUT_IS_LIST), so ComfyUI passes it the whole collection at once rather than item-by-item. That's what makes it fast on big batches.

When you'd actually use it

• Load a folder of character refs with a batch image loader, slice out the first three, and send each to its own ControlNet or IPAdapter branch.
• Take a list of prompts from a text-list builder and split off the first few to test individually.
• Slice a batch of latents from a batch sampler when you only want to keep processing a subset.

Install

Part of AK Pack, so the standard drill applies. In ComfyUI Manager, search "AK Pack" and install; or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/akawana/ComfyUI-AK-Pack

Restart ComfyUI. No dependencies, no models - it's a pure-Python list utility.

Gotchas

• Don't wire all 50 outputs. You'll just be feeding None into half your graph and wondering why things break. Wire what you set length to.
• length caps at 50. If you genuinely need more items out, chain a second node or reconsider the workflow - 50 parallel branches is already a lot.
• Watch the index math. starting_index is offset into the list for item_0. Off-by-one here is the single most common mistake, and the node won't complain - it'll just hand you the wrong image.
• If a target expects a list further downstream and you've sliced it down to scalars, that's the expected trade-off - use a list-join or re-batch node to go back up.