Boyo Load Image List

The complement to Boyo Load Image Batch. Where that node hands you one image at a time to step through, Boyo Load Image List dumps an entire directory into the graph as a single batched tensor, resized to a uniform target size. It's the node for "process all of these the same way" jobs: batch img2img over a folder, batch upscaling, or prepping a set of reference images for a dataset. Point it at a directory, tell it the target dimensions, and out comes one IMAGE tensor holding every frame plus a comma-joined string of the filenames.

It reads .png, .jpg, .jpeg, .webp, and .bmp, sorted alphabetically, so the batch order is predictable and the filenames output lines up with the frame order.

The inputs that matter

• directory - relative path, default input_images. Note this resolves relative to where ComfyUI runs, so the conventional spot is ComfyUI/input/input_images/.
• target_width / target_height - the uniform size every image gets forced to, defaults 512×512. Choose what your model actually wants.
• resize_method - how images are forced to fit:
  • resize - plain stretch to the target. Fast, but distorts aspect ratios.
  • crop - center-crop to the target aspect, then resize. No distortion, loses edges.
  • pad - fit inside the target aspect and fill the letterbox with pad_color. No distortion, no content loss.
• pad_color - fill color for the pad method, default black. Accepts named colors.

Outputs: images (the full batch) and filenames (a comma-separated string, e.g. a.png,b.png,c.png).

Install

Part of Boyonodes. ComfyUI Manager → search "Boyonodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/DragonDiffusionbyBoyo/Boyonodes

Restart. No extra pip packages - plain PIL.

Gotchas

It throws if the directory doesn't exist or contains no valid images, so the error is loud, not silent. And because everything gets forced to one size, mixed-ratio folders will either stretch (resize), lose content (crop), or gain bars (pad) - pick the method that matches your downstream use. If you're feeding a dataset trainer that wants square crops, crop is usually right; if you're just batch-generating previews, pad preserves everything.