Random Mixer🧸

Random Mixer takes up to five text inputs and randomly reshuffles which output each one comes out of. You feed it Text_1 through Text_5, it picks a permutation based on a seed, and the same strings come back out on Output_1 through Output_5 - just in a scrambled order. That's the whole trick. It's a randomizer for wiring, not for content: nothing gets rewritten, only reordered.

Why would you want that? Say you've got five prompt fragments, five style tags, or five caption variants, and downstream you've got five slots that consume them. Instead of manually rewiring every run to try different combinations, you let Random Mixer roll the assignment for you. Change the seed, get a different pairing. It's the kind of thing you reach for when you're batch-exploring variations and you want the machine to stop you from unconsciously always putting your favorite input in the same place.

This is one of the small utility nodes riding along in glibsonoran's Plush pack. The headline acts in Plush are the LLM-driven ones - Style Prompt and Advanced Prompt Enhancer, which call out to ChatGPT, Groq, Anthropic or a local model to write prompts for you. Random Mixer is the opposite of that: no API key, no model, no network. It's plumbing that got added in the October 2024 batch of "randomizing utility nodes" alongside its sibling, Random Output.

How it works

Everything hinges on the seed. Random Mixer uses it to deterministically pick which input lands on which output. Same seed, same shuffle, every single queue - so if you want a new arrangement each run, you have to actually change the seed. In the ComfyUI UI that means setting the seed control to randomize (or increment) rather than fixed. Leave it fixed and the node happily hands you the identical permutation forever, which trips people up: it looks broken because "the random node isn't changing anything," when really it's just doing exactly what a fixed seed tells it to.

Inputs and outputs

Per the node's schema, the inputs and outputs that matter are:

• Text_1 … Text_5 - five optional multiline string inputs. Connect as many or as few as you need; you don't have to fill all five.
• seed - an integer (default 9). This is the only real setting. It selects the shuffle.
• Output_1 … Output_5 - five string outputs carrying your inputs, reordered.

There's nothing else to configure. Wire your strings in, wire the outputs to wherever they're consumed, and drive the seed.

Installing it

ComfyUI Manager is the easy path: open the Manager, hit Install Custom Nodes, search Plush, find Plush-for-ComfyUI, install, restart. Or clone it by hand:

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

Then restart ComfyUI. The requirements.txt step pulls the OpenAI Python library (1.3.5+) that Plush's LLM nodes need - Random Mixer itself doesn't touch it, but the install script runs for the whole pack, so let it finish. No model files to download for this node.

Common issues

The one that bites everybody: "my random node produces the same result every time." That's the fixed seed. Flip the seed control to randomize and it re-rolls per queue. It's the same gotcha Plush's own Dall-e node documents about its mock seed value - a fixed seed means run-once, deterministic behavior by design.

Second, remember it's a reorderer, not a generator. If two of your inputs are identical strings, shuffling them is invisible; you need genuinely different inputs to see the effect. And if the pack fails to load at all, that's almost always the pip install choking on the OpenAI dependency rather than anything to do with this node - check ComfyUI's startup log, resolve the dependency, restart. As one of the pure quality-of-life utilities in the pack, Random Mixer has essentially nothing of its own to go wrong.