Load Image List

Loading several images into one ComfyUI workflow is where the "one image in, one image out" habit starts to fail. Core ComfyUI's Load Image hands you a single image, and the moment you feed multiple references downstream you're stacking them into a batch or loading them one node at a time. SimpleSyrup.LoadImageList is the fix for the specific case that matters in regional workflows: it loads files in selection order and hands you an IMAGE list, where each item keeps its own dimensions.

That last bit is the whole point. ComfyUI's IMAGE type is usually a single tensor with a batch dimension, which silently assumes every frame is the same size. The moment your reference images differ - a 1024 portrait and a 1536 landscape, say - a plain batch either forces a resize or errors out. This node returns a native list instead, so each item stays independent. "Loads an ordered image list without resizing or batching its items," as the pack puts it, and that's a stronger promise than it looks like.

Why order matters: SimpleSyrup's regional prompt machinery is order-based. The pack was written by someone who missed ADetailer's inline [SEP] prompt batches, where prompt number one is the global prompt and each prompt after that binds to the corresponding region - first region gets the first prompt, and so on. If you're feeding reference images or masks into that pattern, the order you pick in the loader is the order everything downstream assumes. Select them deliberately; re-selecting is a two-second fix, but an out-of-order image in a region workflow produces the same class of "why is the wrong character wearing the wrong clothes" bug that ADetailer users hit with [SEP].

The interface is a multi-select combo with an upload button (it's a ComfyUI v3 MultiCombo input, with the chip UI). You pick files in order, each becomes one IMAGE list item, and the images output is that list. Wire it to anything that accepts an image list - reference-image conditioning, the pack's regional nodes, or an IPAdapter-style stack. One nice property: because it's a list output rather than a batch, downstream nodes that genuinely need a single tensor get one item at a time via the usual list-handling, and nothing is silently resized on the way in.

Install: this is one of twenty-ish nodes in the SimpleSyrup pack, so you install the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Artificial-Sweetener/SimpleSyrup
cd SimpleSyrup && pip install -r requirements.txt

or, easier, ComfyUI Manager → search SimpleSyrup → Install → restart. SimpleSyrup uses ComfyUI's v3 extension API, so a reasonably current ComfyUI is required. No heavy models ride along with this node - it's a loader for files you already have, so the only failure mode worth knowing is the obvious one: a missing file shows up as an error at load time, and the v3 schema validates your selection before executing, which is friendlier than the old silent-fail behavior.