RloadNpyInput

RloadInput is Rload with one extra input - anything - and that input is, frankly, a decoy. Look at the code: the node accepts it and then throws it away, overwriting it with whatever the .npy load produced. So why does the input exist at all?

Two genuinely useful reasons. First, splicing: if a live branch's output feeds a bunch of downstream nodes and you want to serve a saved snapshot in its place, drop RloadInput onto that wire, connect the old output to anything, and every downstream connection stays put while the node now outputs the loaded data. You swap a snapshot in without re-drawing half the graph. Second, ordering: a wired input forces the node to evaluate after whatever feeds it, which is what you want when the load must happen behind an upstream run.

How it works

Otherwise it's identical to Rload (displayed as RloadNpyInput). file_path and file_name (default ComfyUI_00001_) concatenate and resolve against ComfyUI's output directory, so the default reads ComfyUI/output/ComfyUI_00001_.npy. The value comes back through output (*), as a batch-dimensioned torch tensor if the first element is tensor-shaped, raw otherwise.

Inputs and outputs

• anything (*) - the splice input. Whatever you feed it is discarded, not merged. Its only job is to hold the wire you're replacing.
• file_path, file_name - same as Rload. Type the prefix the save side used.
• output (*) - the loaded data.

Install

Same as every node in the pack - no deps, no models. ComfyUI Manager → search ComfyUI-ResourcesLoad → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MrFrankHobbidy/ComfyUI-ResourcesLoad

Restart and you're set.

The trap to internalize

It is not a pass-through and it is not a merge. Wire something into anything and you get the loaded file's contents back, not a combination - the input value never survives. If you expect it to, you'll be confused for exactly one session. Also inherit Rload's caveats: the .npy must exist with the right prefix, there's no IS_CHANGED so a stale cached value can linger if nothing else changes, and allow_pickle=True means you should only load files you trust.