📸 Multi Images Input v2

What it does

Combines up to sixteen IMAGE inputs into one batched IMAGE tensor. That's the whole node. Its job in this pack is feeding the multimodal node a set of stills - reference sheets, comparison shots, a storyboard - without stringing six Load Image nodes through reroutes to build a batch by hand. It works anywhere a batch is expected, not just with the Gemini nodes.

How it works

inputcount says how many slots to read. The node walks image_1 through image_<inputcount>, skips the ones you left empty, flattens any input that's already a batch frame-by-frame, then normalises everything: images that don't match the first image's height and width get resized with a bilinear interpolate, values are clamped to 0–1, and the lot is stacked into one tensor.

Inputs and outputs

Required is just inputcount (1–16, default 4). Optional are image_1 … image_16. One output: images.

What bites people

inputcount is a ceiling, not a hint. Wire something into image_7 while inputcount sits at 4 and that image is silently discarded. No warning, no error, just a missing frame in the batch. Set the count to match your highest wired slot.

Everything gets squashed to the first image's geometry. The resize is a plain stretch, not a crop or a pad. Feed it a 16:9 and a 9:16 and one of them comes out distorted. If the aspect ratio matters, put the image whose shape you want in image_1 - that's the template.

Order is slot order, not wiring order. The batch comes out 1, 2, 3… regardless of which cable you dragged first. When the model is comparing images, "the second one" means image_2.

Feeding it nothing returns a black frame, not an error. With no inputs connected the node hands back a single 512×512 black image, which sails into the API node as a valid image and produces a confident analysis of nothing at all. If your output reads like the model is describing a void, check what's actually plugged in.

Install

Same pack, same routine - Manager search for ComfyUI Gemini 3x Pro, or:

cd ComfyUI/custom_nodes
git clone https://github.com/asirusasr-maker/ComfyUI-Gemini_3x_Pro

Then install its dependencies with the interpreter that runs ComfyUI and restart:

python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-Gemini_3x_Pro\requirements.txt

No API key needed for this one. It's pure tensor plumbing on your own machine - the only node in the pack that never touches the network.