Random Output🧸

Random Output takes a single text input and randomly routes it to one of several output sockets. You give it a string, tell it how many outputs are in play, and it picks one - seeded - to send the text down. The other outputs sit empty. Think of it as a random one-to-many switch: same value in, unpredictable exit door.

That's genuinely useful when you want a workflow to branch differently from run to run without you touching it. Route a prompt to one of three different downstream chains, send a caption to a random destination, or scatter a value across a batch so different queues take different paths. It's the "randomly present this to one node out of a group" idea, done for text. If Random Mixer (its sibling in this pack) shuffles many inputs across many outputs, Random Output is the simpler cousin: one input, one randomly chosen output.

It comes from glibsonoran's Plush pack, which is really built around the flashy stuff - Style Prompt and Advanced Prompt Enhancer, the nodes that call ChatGPT, Groq, Anthropic or a local LLM to generate prompts for you. Random Output is one of the little Plush/Utils helpers that need none of that: no API key, no model, no network call. It landed in the October 2024 batch of randomizing utility nodes and does exactly one small thing.

How it works

The seed decides everything. Given a seed and a number of active outputs, Random Output deterministically picks which output receives your text. Same seed, same choice, every queue - so to actually get randomness across runs you set the seed control to randomize (or increment) instead of fixed. This is the single most common source of "it's not random" confusion, and it's not a bug: a fixed seed is meant to be repeatable.

The count of outputs in play is controlled by a separate setting, so you're not stuck routing across all five - you can tell it to choose between just the first two, or three, and leave the rest out of the draw.

Inputs and outputs

From the node's schema, the ones you actually set:

• Text - a multiline string input (required). The value you want routed.
• randomized_outputs - a menu of 1 through 5 (default 2). How many output slots the random pick chooses among. Set it to 3 and the text lands on one of the first three outputs; the rest stay empty.
• seed - an integer (default 9). Selects which output wins the draw.
• Text_1 … Text_5 - the five string outputs. On any given run, one of the enabled ones carries your text and the others are empty.

Downstream, wire each output to whatever chain should run when the draw lands there. Nodes fed by an empty output simply have nothing to do.

Installing it

ComfyUI Manager: Install Custom Nodes, search Plush, pick Plush-for-ComfyUI, install, restart. Manual clone if you prefer:

cd ComfyUI/custom_nodes
git clone https://github.com/glibsonoran/Plush-for-ComfyUI.git
cd Plush-for-ComfyUI
python -m pip install -r requirements.txt

Restart ComfyUI afterward. The requirements.txt install fetches the OpenAI library that Plush's LLM nodes depend on; Random Output doesn't use it, but let the install run to completion so the whole pack loads. No models needed.

Common issues

The recurring one is the seed again: leave it on fixed and the "random" node always picks the same output. Set it to randomize to re-roll each queue. Predictable per seed is the intended behavior.

The other thing worth internalizing: the non-selected outputs emit nothing, not an empty-but-valid value you can rely on. If a downstream node expects a string and gets an unfed input, it just doesn't execute on that branch - which is usually what you want, but it does mean you can't assume all five branches fire. And if the pack won't load at all, the culprit is almost always the OpenAI dependency install failing at startup, not this node - check the ComfyUI console, fix the dependency, restart. Beyond that there's very little surface area here; it's a router, and a small one.