DJZ Zero ENH V1

DJZ Zero ENH V1 is the original version of this pack: it takes your prompt, detects which semantic categories are missing (subject, environment, lighting, camera, style, mood…), and fills those gaps from a hardcoded default vocabulary - no LLM, no API, no profile files. Same (prompt, seed, intensity) gives you the exact same enhanced prompt every time, on any machine. It's the proof-of-concept that the ZeroBytes "position-is-seed" hashing trick works before the pack grew themes.

If you just want a prompt padded deterministically and don't care about genre vocabularies, V1 is honestly enough. Its bigger sibling V2 with the default.json profile is behaviorally the same thing - V1 is literally described in the repo as the hardcoded vocabulary that default.json in V2 replicates. The only real reason to pick V1 over V2 is if you want the leaner node with no profile dropdown in the way, or you're intentionally running an old workflow that predates V2.

How it works

Short version: tokenize your prompt, classify tokens into eight categories, find the gaps, then pick each fill word with an xxhash of (seed, input_hash, coordinate) mod the pool size. A pure hash function can't be random or stateful, so the same coordinates always return the same vocabulary entry - that's the entire determinism guarantee. It applies anti-pairing rules so an underwater scene won't be served golden hour lighting, merges your original tokens, dedupes, and truncates to your word cap. LoRA trigger words and other unclassified tokens pass through untouched.

The inputs that matter

• prompt - the text to enhance (multiline).
• seed - the reproducibility knob; change it to get a different-but-still-deterministic enhancement. Range is 0 to 4294967295.
• intensity - minimal (25%) / light (50%) / moderate (75%, default) / full (100%) of detected gaps filled.

Optionally max_words (default 150, min 10, max 500), plus prefix and suffix strings. The single output, enhanced_prompt, feeds straight into your positive conditioning - same wiring as any prompt text node.

Install

ComfyUI Manager: search "ZeroENH" and install. Or:

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

Then restart ComfyUI. The dependency is a single line - xxhash>=3.0.0 - and there are no model files at all. It's under the DJZ-Nodes category.

The honest take

Keep max_words in check. The attention cap on modern encoders means a 150-word padded prompt can start hurting adherence - if your input is already detailed, reach for minimal or light rather than full. And know the context: this is a small, quiet pack (zero reddit footprint; the author's DJZ repos have even 404'd during a reorganization in mid-2025), so it's a niche utility, not a community staple. For the job it does - zero-cost, offline, reproducible prompt enhancement - it works exactly as advertised.