🖼️ 多圖輸入 / Multi Image Input

The 🖼️ 多圖輸入 / Multi Image Input node is the little helper that makes the pack's Universal Generator actually useful. Nano Banana Pro can take up to 14 reference images in one call, and Veo and Sora happily condition on reference frames too - but plumbing a dozen images into a ComfyUI node is normally a chore. This node turns it into a paste.

The trick it pulls off that vanilla ComfyUI can't: normal IMAGE batches require every frame to be the same size, which is why reference-image workflows get fiddly fast. This node keeps each image its own dimensions. That 4K photo and that square crop ride to the API together in one REPLICATE_IMAGE_LIST, no resizing dance required.

How to use it

Paste one file path per line into image_paths - absolute paths, or paths relative to your ComfyUI input folder. Windows "Copy path" quotes get stripped automatically, and EXIF orientation is applied, so phone photos load the right way up:

D:\photos\ref1.jpg
"C:\Users\me\Pictures\ref 2.png"
my_input_image.png

The image_list output plugs straight into the Universal Generator's images socket. If 14 images isn't enough (or you've already got some wired into the graph), there's image_1 through image_6 for direct IMAGE connections, and an image_list input that chains another Multi Image Input node onto the front. Merge order is image_list → image_1–image_6 → image_paths.

One nice detail: the node implements IS_CHANGED, so if you edit a reference file on disk and hit Run again, ComfyUI reloads it instead of caching the old one. The path parser is also forgiving - a bad path prints a warning and is skipped rather than silently corrupting your call.

Install and gotchas

It ships in the same pack as the generator, so installing that gives you this one too - ComfyUI Manager, search Replicate API NM, or:

cd ComfyUI/custom_nodes
git clone https://github.com/neurmostudio0409/ComfyUI-replicate-api-NM
cd ComfyUI-replicate-api-NM
pip install -r requirements.txt

There's nothing else to it - no API key of its own, no heavy dependencies, nothing to configure. The one trap is the same one as every path-based node: keep paths absolute or input-dir-relative, because ComfyUI's own filepicker outputs resolve differently depending on where you clicked. If an image isn't loading, that's the first thing to check.