Zenkai Prompt V3

Zenkai Prompt V3 is the family's remix machine. Where the plain Zenkai-Prompt just returns a random line from a text file, V3 takes that line, chops it into sections, and swaps some of those sections with parts of other lines - from the same file, or from completely different files in the pack. The result is a mashup prompt you'd never have written by hand. Think of it as "interpolation" in the literary sense: you're blending sources, not math.

It's part of the Zenkai prompt family in DJZ-Nodes, and the use case is exploration. You want weird, unexpected combinations to feed a batch, or you're stuck and want the node to surprise you with a premise. It won't always be coherent - that's the point, and you'll want to prune the output before committing to a final render.

How it works

The mechanism is clearer than the name suggests. The node:

1. Picks a base line from your chosen file using the seed.
2. Splits that line into N sections by word count (N comes from division_preset).
3. Depending on interpolation_mode, replaces every section after the first:
   • lines - each section gets swapped with the matching section of a random other line from the same file.
   • directory - sections get swapped with matching sections of lines pulled from random files across the whole prompts folder.
4. Joins the sections back together and wraps them in your optional prefix/suffix.

So with fifths + directory, you get a sentence whose opening comes from one story, whose middle comes from a cyberpunk file, and whose ending comes from a third source entirely. It's a cut-up experiment, Burroughs-style, running inside your node graph.

The inputs that matter

The seed and text_file behave like the base node. The two that make V3 V3:

• interpolation_mode - none (just return the base line, effectively the same as the plain node), lines, or directory. This is the flavor dial.
• division_preset - words through fifths. Controls how many sections the prompt is carved into, and therefore how much of it gets replaced. More sections = wilder blends. With words (split every word), the result is nearly random text.

Plus the optional prefix and suffix strings, which wrap whatever came out.

Output

One STRING - the interpolated line. Feed it into a CLIP Text Encode and see what your model makes of the mashup.

Install

Same pack as everything else here:

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/DJZ-Nodes
cd DJZ-Nodes
pip install -r requirements.txt

Restart, or use ComfyUI Manager ("DJZ-Nodes"). Pure-stdlib node; the pack's heavy dependencies are for its other 100+ nodes, not this one.

Gotchas

• Results are chaotic on purpose. Directory mode pulling from 204 files will occasionally produce nonsense. That's the feature - treat output as raw material.
• The split is by word count, not by sentence or clause, so sections can cut mid-phrase. Accept the jank.
• Seed does double duty: it both picks the base line and drives the section swaps, so a new seed is a fully new blend.
• Prefix/suffix only wrap the final string - they don't get interpolated themselves.

If you want reproducible, curated randomness from a single file, the plain Zenkai-Prompt is calmer. If you want the node to actively surprise you, V3 is where that starts.