Image Selector (ae)

Image Selector (ae) is a window that pops open when you run the workflow, shows a folder's images as a thumbnail grid, and hands your picks to the next node as a proper IMAGE batch. If you've ever dumped a whole folder into a loader and hoped, or cursed a script that can't filter out the five blurry shots, this is the human-in-the-loop middle ground.

Why you'd reach for it: curation. ComfyUI's whole philosophy is that every step is a node and data flows between them, but "which of these 200 images do I actually want" is a judgment call, not a computation. This node turns that judgment into a click. The classic use is feeding a curated selection into a training-style pipeline - and in this same pack, the repo's example workflow wires ImageSelector straight into the Z-Image Images To LoRA node. You browse, click the good references, and the LoRA node only ever sees what you chose.

How it works: it's a pygame window - the same interactive engine as the pack's pose editor - so it blocks the queue while it's open. You select with clicks, then hit ESC or ENTER and the batch flows downstream. The smart part is loading: folders load asynchronously, thumbnails render lazily with caching, only visible images stay in memory, and there's a one-image-per-frame cap. A folder with thousands of images scrolls instead of freezing. Your selection order is preserved in the output tensor, which matters if the next node cares about ordering.

Inputs you actually set: folder_path is required - a starting directory, and you can navigate the folder tree from inside the window, so it doesn't need to be the exact target. target_width and target_height (0 = auto) force consistent output sizing, handy when the next node wants a uniform batch. seed does the pack's usual job - fixed means it runs once, change it to open the browser again. Outputs are images (the selected batch in ComfyUI IMAGE format) and masks (alpha masks for images with transparency, useful when you're working with cut-outs).

Install is painless compared to the pack's other star. pygame ships in the pack's requirements.txt, so it's the standard route: ComfyUI Manager → Install via Git URL (https://github.com/by-ae/ae-in-workflow.git) → restart, or clone into custom_nodes and pip install -r requirements.txt. No model downloads, nothing heavy.

The two real gotchas. First, this needs a display: if ComfyUI runs headless on a server or in a container you SSH into, the window won't appear. It's a local-machine node, full stop. Second, it's interactive by design - the pack documents that these editor nodes require a human, so it's wrong for fully automated batch jobs. A couple of smaller ones: a folder path that doesn't exist raises an error (double-check it), and an empty folder returns an empty batch rather than an error. If you want a different folder, don't restart - just navigate with the tree panel.