CRZ Image Selector

CRZ Image Selector is the one dashboard node that actually touches pixels. It gives you six image slots - each one shows a thumbnail you can click to pick a file - and then hands you back the loaded images as tensor outputs, an image count, and a combined batch. It's the pack's answer to "I want to feed two or three reference images into my workflow without hunting through a Load Image node for each one."

It's from ComfyUI-CRZnodes, CoreyCorza's dashboard pack, and it's the heaviest node in the family: it's the only one doing real work on the backend (loading, EXIF handling, resizing, batch concatenation) rather than just passing values through.

How it works

The six inputs are plain file paths (strings) relative to ComfyUI's input directory - the same place your Load Image node looks. On the frontend, each slot draws a thumbnail and a clear button; clicking a thumbnail opens ComfyUI's file dialog to pick (or re-pick) an image. On the backend, each path is loaded with PIL, EXIF-transposed, converted to RGB, and normalized to a tensor.

Two behaviors are worth knowing about. First, empty slots return a blank 64×64 black image, not nothing - so if you leave slot 3 empty, its image_3 output is a tiny black tensor, and you should read image_count to know how many real images you actually have. Second, the image_batch output stacks only the non-empty images into a batch, and if the source images have different dimensions it resizes them (lanczos, center-cropped) to match the first image's size. Sizes will be silently normalized.

The inputs and outputs that matter

• image_1 … image_6 (inputs, STRING) - file paths in ComfyUI's input folder. Usually you don't type these; you click the thumbnails.
• image_1 … image_6 (outputs, IMAGE) - one tensor per slot (blank if the slot is empty).
• image_count (output, INT) - how many of the six slots are actually filled. Read this before trusting the batch.
• image_batch (output, IMAGE) - all filled images concatenated into a single batch, size-normalized.

Six inputs, eight outputs. The two you'll actually use are the individual image you care about and image_batch when you want a stack for batch processing.

Installing it

Standard pack install: ComfyUI Manager → search ComfyUI-CRZnodes → install → restart. No requirements beyond what ComfyUI itself ships - PIL, numpy, and torch are all core ComfyUI dependencies, so there's nothing extra to install.

cd ComfyUI/custom_nodes
git clone https://github.com/CoreyCorza/ComfyUI-CRZnodes.git

Gotchas

The blank-image-for-empty-slot behavior is the trap. It means the individual outputs always fire, so you can't use "is there a connection" to detect whether a slot is used - use image_count for that. It also means if you wire an empty slot's output into a detailer or a ControlNet, you'll be feeding it a 64×64 black image, which will absolutely change your result. Check image_count in any logic that branches on how many images you're using.

The batch resizing is the second one to know: mixed dimensions get silently squashed to the first image's size. If your reference images have different aspect ratios, expect cropping. For most reference-image use that's fine, but if exact framing matters, normalize your source images first.