ADV Image Series (4)

If you've ever tried to bundle a few images of different resolutions onto one wire and had ComfyUI silently refuse, you already know why this node exists. ADV Image Series (4) takes up to four IMAGE inputs and emits them as a list - so the images that can't share a single tensor can still share a single output port.

It's the image half of the pmarmotte2/Comfyui_Pick_Any pack, designed to feed the pack's ADV Pick Any Chooser. Feed it your four candidate frames, wire its images output into the chooser's images port, and you get a picker dialog full of thumbnails instead of a pile of loose Save Image nodes.

How it works

ComfyUI's IMAGE type is a batched tensor - one wire, one stack of frames, all the same size. That's great until you want to offer up a 512x512, a 768x768, and a 1024x576 as options. Stick them in a normal batch and anything downstream that does tensor math errors out on the mismatched shapes.

This node sidesteps that by not batching at all. It takes whatever each of its four slots receives, normalizes each to a clean single-frame tensor (stripping any stray singleton dimensions), and emits a Python list of images instead of a tensor. Different resolutions coexist happily, because nothing is ever stacked together.

Inputs and outputs

• image_1 through image_4 - all optional, all IMAGE. Leave slots empty and they're simply skipped; the node drops anything None before building the list.
• images - the single output, a list of images. It feeds the chooser's images port directly, or any node that knows how to consume an image list.

Note this is the only image-series node in the pack - there's no 8-input variant like the audio and text series get. Four slots is the whole deal.

Install

It ships with the Comfyui_Pick_Any pack, so install once:

cd ComfyUI/custom_nodes
git clone https://github.com/pmarmotte2/Comfyui_Pick_Any

Restart ComfyUI and hard-refresh the browser tab. No extra Python dependencies, no model files - this is a pure utility node, which means it also can't be the thing that breaks your environment. Folder name without spaces, per the README.

Common issues

• Nothing comes out when slots are empty. Expected. The node filters empties and None inputs, so an all-empty node emits an empty list, and the chooser just shows nothing to pick. That's fine - the inputs are there so you can build a workflow where slots fill up as generations complete.
• The downstream node rejects the list. A plain list won't plug into nodes expecting a batched tensor. That's by design: route lists into the chooser or another list-aware node, not into something expecting a single IMAGE.
• "It doesn't resize my images." Correct, and the README's "mixed image sizes" claim means it tolerates them, not that it fixes them. If you need everything at one resolution, upscale/resize before this node.