Random Vocab String List (Nukun)

"Random prompt" nodes usually disappoint in one of two ways: they repeat the same handful of words because sampling from a small list with replacement is a small world, or they're genuinely random and you can't reproduce a good run. This node dodges both with a shuffle-bag approach. It reads a vocabulary file, shuffles it deterministically, and hands you a slice - increment the seed and you walk through non-overlapping blocks, so ten queued runs cover ten different corners of the vocabulary before anything repeats. Reproducible and varied at the same time. That's a real feature, not a gimmick.

The intended role in this pack is as the front-end for the Ollama Prompt Refiner: random words in, curated split prompts out. On its own, though, it's a perfectly good way to add controlled randomness to any workflow that wants a text input.

How it works

Pick a vocab_file from the dropdown - the pack bundles 30+ files covering places, objects, animals, verbs, nouns, adjectives, *ing words, camera/composition terms, quality tags, person names, countries, cities, and 250 visual art styles, plus little_doom_*.csv files with LoRA keyword subsets. It also reads ComfyUI/user/vocab.json if you have one. Words are deduplicated before sampling, and amount gets clamped to the vocabulary size, so asking for 50 words from a 12-word file gives you 12, not an error.

The seed widget is the cursor. With control-after-generate on increment, each queued run moves to the next shuffled block; randomize jumps to another block. Want to grow a prompt? Connect the optional chain input and generated words append after it. That's the whole interface: vocab_file, amount, seed, optional chain, one STRING output.

The formats are plain comma-separated text. The bundled guide warns about the one parsing gotcha: no leading spaces after commas (tag_a,tag_b not tag_a, tag_b), and the same applies to any custom files you drop into custom_nodes/Nukun_ComfyUI_Nodes/resources/. .csv, .txt, and .json all work.

Where it sits in a workflow

Straightforward: output → prompt encoder, or output → Ollama Prompt Refiner's word_salad for the full pipeline. If one file isn't enough variety, the pack's Multi Vocab String List (Nukun) combines four vocab files with independent amount and word_index cursors per slot.

Installing it

Nothing special - it's just text parsing:

cd ComfyUI/custom_nodes
git clone https://github.com/OnekoSL/Nukun_ComfyUI_Nodes.git

Restart ComfyUI, or install via Manager (search "Nukun"). No model downloads, no dependencies beyond the pack's own.

Common issues

The beginner trap is expecting the seed to act like a normal random seed. It doesn't - it's a block cursor, so incrementing by 1 gives you a different block, but incrementing by 50 might wrap you back near the same block you saw before. If your random prompts feel less random than you'd like, it's because a shuffle-bag guarantees coverage over variety; that's the trade you're buying reproducibility with. Also check your vocabulary file if you're seeing repeats across the whole range - a short list reshuffles fast no matter how clever the bag is. And if you add a custom file and it doesn't show in the dropdown, restart ComfyUI; the node scans resources/ at load time.