Multiple Images Input

Multiple Images Input does one job: it takes any number of separate IMAGE inputs and stacks them into a single IMAGE batch. On its own that's not exciting - it's plumbing. What makes it worth an article is what it exists to unblock in this pack: MiniCPM_VQA_Polished, the vision-language Q&A node next to it, only has room for one source_image slot. If you want to ask a question that spans four, six, or twenty images at once - "what's changed across these frames" or "describe this whole set" - you can't wire them all in directly. You batch them here first, then feed the single images output into that one slot.

The alternative, the older MiniCPM_VQA node, sidesteps the need for this by hard-coding three named image sockets instead. That's fine for exactly two or three images and nothing else - but it's a hard ceiling. Multiple Images Input is the node that removes the ceiling: it scales to as many images as the widget will let you add (up to 1000, per the node's own limit), which is the actual reason to reach for the Polished VQA node over the legacy one once you're past a handful of images.

How it works

The interface is a little different from a normal ComfyUI node because the number of inputs isn't fixed - it grows. You start with two required sockets, image_1 and image_2, plus an inputcount widget (default 2, minimum 2, max 1000). Type a bigger number into inputcount and click the node's update control, and it adds however many new numbered image_N sockets you asked for. Wire an image into each one you've added.

Once everything's connected, the node concatenates them into a single images output - one IMAGE tensor carrying every image you fed in, in order. That's what makes it compatible with anything downstream expecting one IMAGE input rather than a list: MiniCPM_VQA_Polished's source_image slot, most obviously, but really any node in your graph that's happy to receive a batch.

The inputs and output that matter

• inputcount (INT, default 2, min 2, max 1000) - set this to however many images you actually want, then trigger the update. It's the only control on the node itself; everything else is just the image sockets it generates.
• image_1, image_2, ... (IMAGE) - one socket per image, added as you raise inputcount. There's no cap you'll realistically hit short of it.
• images (IMAGE, output) - the batched result. Wire this straight into MiniCPM_VQA_Polished's source_image input.

Installing it

It ships as part of the same pack as the MiniCPM VQA nodes, so there's no separate install step. Via ComfyUI Manager: search minicpm, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/IuvenisSapiens/ComfyUI_MiniCPM-V-4_5
pip install -r ComfyUI_MiniCPM-V-4_5/requirements.txt

then restart. This particular node doesn't touch any model weights or heavy dependencies itself - it's pure tensor plumbing - so if it's not showing up after install, the problem is almost certainly the pack failing to load rather than anything specific to this node. Check your ComfyUI console output on startup for an import error from the pack.

Common issues & troubleshooting

Raised inputcount but no new sockets appeared. You need to actually trigger the update after changing the number - just editing the widget value isn't enough on its own for a dynamic-input node like this one; look for the update control on the node body.

Images come out the wrong resolution or the batch errors downstream. Batching assumes the images are compatible sizes going in - if your sources are different resolutions, resize them to match before this node, the same way you'd need to for any other IMAGE batch operation in ComfyUI.

You only need two or three images. This node is solving a scaling problem you might not have. If you're always working with exactly two or three fixed images, the legacy MiniCPM_VQA node's three named sockets do the same job with one fewer node in the graph - reach for this one specifically when the count varies or climbs past what fixed sockets can handle.