Repeater 🐍

The README's own example is the clearest way into this node: point Repeater at a Checkpoint Loader and you get N copies of that same loaded model, useful for wiring the same checkpoint into several parallel branches. Point it at a random-seed generator instead, and you'd want N different values, not the same seed copied five times. Repeater handles both cases - the difference is one setting.

It's part of ComfyUI-Custom-Scripts, pythongosssss's utility pack, and it's honestly labeled [WIP] in the README - worth knowing going in, more on that below.

How it works

source is the wildcard input - wire in whatever node or value you want multiplied. repeats sets how many copies, from 0 to 5000 (default 2). output picks the shape of the result: single gives you one list-typed output carrying all the repeats, multi gives you N separate output sockets on the node itself, letting you wire each copy to a different downstream slot directly.

node_mode is the setting that answers the checkpoint-vs-random-seed question above: reuse evaluates the source once and pipes that same result to all N slots (correct for anything where re-running would just waste compute on an identical result, like a checkpoint load); create actually re-executes a fresh instance of the source node for each copy (correct for anything that should genuinely differ per copy, like a random value generator).

The honest caveat

The README states this plainly, and it's better to know it going in than to discover it mid-workflow: Repeater is marked [WIP], with a standing TODO for type safety on the wildcard outputs. In practice that means the node won't stop you wiring a repeated output into a slot expecting a different type - it'll let the connection through, and you'll only find out it was wrong when the workflow actually executes. Treat this as "does the job, rough edges included," not a polished piece to build a production pipeline around unsupervised.

Where it's useful beyond the README's example

Fanning a fixed prompt/seed pair out to several parallel sampler branches for a quick A/B comparison grid. Repeating a loaded image across N separate processing chains that each do something different to the same source picture. Anywhere you'd otherwise be copy-pasting the same node five times by hand.

Installing it

• ComfyUI Manager - search "ComfyUI-Custom-Scripts", install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/pythongosssss/ComfyUI-Custom-Scripts, restart.

No models, no extra dependencies.

Common issues

Downstream node throws a type error after wiring a repeated output. This is the documented lack of type-checking on the wildcard outputs, not a mystery - double-check what type you actually wired in and whether it matches what the receiving node expects.

Getting five identical values when you wanted five different ones. You're on node_mode: reuse. Switch to create so the source node actually re-executes for each copy instead of running once and being reused.

Reused a heavy node (like a checkpoint loader) on create by mistake. That re-runs - and reloads - it N times, exactly the wasted compute reuse exists to avoid. Flip it back to reuse for anything expensive that should stay identical across copies.

Socket count on multi doesn't match what you expected. It tracks the repeats widget value. If you change repeats after already wiring things downstream, you'll need to rewire - the socket count follows the current setting, not whatever it was when you first connected things.