Reference Image Loader

Reference Image Loader is the pack's answer to one of the dullest parts of multi-reference workflows: wiring up a row of Load Image nodes and managing eight separate outputs. Here you hit an "⬆ Upload images" button inside the node itself, pick one or several files (up to 8), and get them back as a single IMAGE batch, ready to feed the multi-reference generators in the same pack.

It exists because of how the pack's generators evolved. Simple Image Generator (Multiple) and Story Frame Generator both take a batched reference_images input - one wire, not a slot per image. This loader is the natural companion: upload your references, reorder or remove them in a compact thumbnail table, and connect the single batch output straight in.

What you get

• batch - all loaded images as one IMAGE tensor.
• image_1 through image_8 - individual outputs, available depending on your output mode.
• count - how many images are loaded.

The output_mode select (Single Batch, Separate Images, or Batch + Separate) controls which slots appear. The frontend hides the ones you're not using.

The nice detail is how it handles mixed sizes: different-sized images are combined without cropping or stretching - they're letterboxed onto a consistent canvas, so the batch stays valid as one tensor and nothing gets squashed. That pairs well with the generators' aspect-preserving reference sizing, which is where the multi-reference edit path wants to be.

The restart gotcha (read this first)

The upload button, thumbnail table, and reordering UI are all added by a frontend script (web/zfrnodes.js), not by ComfyUI core. That means this node demands a full ComfyUI restart after you install or update the pack - a browser refresh will not make the upload button appear. If you add the node and it looks like a plain text field with no button, you skipped the restart.

Install

Via ComfyUI Manager (search "ComfyUI-ZFRNodes") or:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/zfrsgtcu/ComfyUI-ZFRNodes.git

Then restart ComfyUI completely. Requirements are just numpy, torch, and Pillow - no model downloads, no API keys. It's a pure input node: it loads, batches, and outputs.

Typical wiring, per the pack README: Reference Image Loader → Simple Image Generator (Multiple) (reference_images), or → Story Frame Generator (reference_images) for identity-locked stories. If you're feeding a contact sheet from Sheet Compositor into the story path, this loader is how it gets in - or you can connect the sheet's IMAGE output directly, since both are IMAGE types. The loader just makes the uploading part pleasant.