NanoB Reference Adapter (Legacy)

Let's get the blunt version out first: you almost certainly should not use this node. "NanoB Reference Adapter (Legacy)" sits in its own NanoGemini/Legacy folder, and the code calls it exactly what it is - a "Legacy compatibility adapter." If you're building a new workflow, walk past it. Its only real job is keeping old saved workflows alive.

What it actually does

Mechanically, almost nothing. One image socket in, one ref_data output out. The entire function body is:

return ({"image": image},)

It wraps your image in a dictionary and hands you a custom NANO_REF_DATA type. No model runs, no API call happens, there are no weight sliders, no settings - nothing to tune because there's nothing left to tune.

The history explains it. This pack (ARC NanoB Gemini) is a fork of comrender's ComfyUI-NanoB-Edit-Gemini, and older versions of that design pushed every reference through a small config node that could carry a weight/influence value alongside the image. When the fork simplified reference handling, the weight concept went away entirely. The pack's own changelog note says it plainly: "Reference Weight/Influence is retained only as a legacy adapter for older workflows." The class stayed registered so graphs that contain it don't break on load, but the adapter no longer sits between your image and the editor.

The part that surprises people

In the current release, nothing in the pack accepts this node's output type. The editor (NanoB Gemini Editor / NanoBEditGemini) takes images into its own image socket and references into a references socket of type NANO_REFS - a different type than the NANO_REF_DATA this adapter emits. So on a fresh install the ref_data output has nowhere to plug in. It's a dangling token that exists so an old workflow JSON still resolves its class name instead of erroring on load.

If you downloaded a workflow that includes it, that's the tell that the workflow predates the pack's reference redesign. Delete the node and rewire the old-fashioned way:

1. Take the image you want to edit into the editor's image socket.
2. Take every donor/reference image into NanoB Reference Stacker.
3. Run the stacker's references output into the editor's references socket.

Installing it (well, the pack)

You can't install just one node from this pack, and you shouldn't need to - it rides along with the whole thing:

cd ComfyUI/custom_nodes
git clone https://github.com/DHan315/ComfyUI-ARC-NanoB-Gemini

or search "ARC NanoB Gemini" in ComfyUI Manager, then restart ComfyUI. There are no pip dependencies (the project declares an empty dependency list) and no model files to download - the pack's real requirement is a Google Gemini API key for the editor node, set via the GEMINI_API_KEY environment variable or pasted into the node.

When you might actually see it

Load any ARC-era NanoB workflow that's been through a few update cycles and you'll find one of these orphans sitting in the corner, disconnected. That's expected, not a bug - the author kept it around precisely so those graphs don't hard-fail. The moment you touch such a workflow, your first move should be deleting this node and replacing it with the stacker. Nothing of value is lost: the modern path is simpler, and it's the one the maintainer actually recommends.