ComfyUI-ZeroENH

Deterministic Prompt Enhancement via Procedural Augmentation

Part of the DJZ-Zerobytes 0(1) procedural collection

• https://github.com/MushroomFleet/ComfyUI-DJZ-ZeroPrompt
• https://github.com/MushroomFleet/ComfyUI-ZeroENH
• https://github.com/MushroomFleet/ComfyUI-ZeroEDIT-nodes

ZeroENH enhances input prompts with contextually appropriate additions using O(1) coordinate hashing. The same (input_prompt, seed, intensity) combination always produces identical enhanced output - everywhere, every time.

Features

• Perfect Reproducibility: Same inputs = same outputs, guaranteed
• Sub-100ms Latency: No LLM inference required
• Zero Cost: Pure local computation
• Full Auditability: Every enhancement selection traceable to hash coordinates
• Intensity Control: Choose enhancement level (minimal/light/moderate/full)
• Anti-Pairing Logic: Prevents semantically incoherent combinations
• LoRA/Embedding Preservation: Unclassified tokens pass through unchanged

Installation

Via ComfyUI Manager (Recommended)

Search for "ZeroENH" in ComfyUI Manager and click Install.

Manual Installation

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes
   

2. Clone this repository:

   git clone https://github.com/MushroomFleet/ComfyUI-ZeroENH.git
   

3. Install dependencies:

   cd ComfyUI-ZeroENH
   pip install -r requirements.txt
   

4. Restart ComfyUI

Nodes

All nodes are found under the DJZ-Nodes category.

DJZ Zero ENH V1

Hardcoded default vocabulary - no external files required.

Inputs:

Input	Type	Description
prompt	STRING	Input prompt to enhance
seed	INT	Seed for deterministic enhancement (0 to 4294967295)
intensity	COMBO	Enhancement level: minimal (25%), light (50%), moderate (75%), full (100%)
max_words	INT	Maximum word count for output (default: 150)
prefix	STRING	Text to prepend to enhanced prompt
suffix	STRING	Text to append to enhanced prompt

Output:

Output	Type	Description
enhanced_prompt	STRING	The deterministically enhanced prompt

---

DJZ Zero ENH V2

Custom JSON profile support with theme-specific vocabularies.

Inputs:

Input	Type	Description
prompt	STRING	Input prompt to enhance
profile	COMBO	Select enhancement profile (auto-discovered from /profiles folder)
seed	INT	Seed for deterministic enhancement (0 to 4294967295)
intensity	COMBO	Enhancement level: minimal (25%), light (50%), moderate (75%), full (100%)
max_words	INT	Maximum word count for output (default: 150)
prefix	STRING	Text to prepend to enhanced prompt
suffix	STRING	Text to append to enhanced prompt

Output:

Output	Type	Description
enhanced_prompt	STRING	The deterministically enhanced prompt

---

DJZ Zero ENH Profile Info

Utility node to display profile statistics.

Inputs:

Input	Type	Description
profile	COMBO	Select profile to inspect

Output:

Output	Type	Description
info	STRING	Formatted profile statistics (pool sizes, combinations, rules)

---

Included Profiles

The following profiles are included in the nodes/profiles/ directory:

Profile	Description
default.json	Universal vocabulary for general image generation (V1-compatible)
cyberpunk.json	Neon-noir dystopian vocabulary (Blade Runner, Ghost in the Shell, Akira aesthetics)

Creating Custom Profiles

Add .json files to the nodes/profiles/ directory. They will be auto-discovered on ComfyUI restart.

Profile structure:

{
  "name": "Profile Name",
  "description": "What this profile does",
  "version": "1.0.0",
  "type": "enhancement",
  "templates": ["..."],
  "pools": {
    "subject": ["..."],
    "action": ["..."],
    "environment": ["..."],
    "style": ["..."],
    "lighting": ["..."],
    "camera": ["..."],
    "details": ["..."],
    "mood": ["..."]
  },
  "classification": {...},
  "rules": {...},
  "anti_pairs": {...}
}

How It Works

ZeroENH uses the ZeroBytes position-is-seed methodology:

1. Tokenization: Split input prompt into semantic tokens
2. Classification: Identify which categories are present (subject, environment, style, lighting, camera, details, mood, action)
3. Gap Finding: Detect missing categories
4. Hash Selection: Use xxhash32(seed + input_hash + coordinate) to select enhancements from vocabulary pools
5. Anti-Pairing: Avoid semantically incompatible combinations
6. Template Assembly: Merge original tokens with enhancements
7. Deduplication: Remove redundant elements
8. Length Enforcement: Truncate if exceeding word limit

Determinism Guarantee

hash(seed, input_hash, coordinate) % pool_size → selection index

This is a pure mathematical function with no state or randomness. The same coordinates always produce the same selection.

Enhancement Categories

Category	Description	Example Keywords
Subject	Main entity/character	woman, dragon, robot
Action	Movement/activity	standing, running, flying
Environment	Setting/location	forest, city, space station
Style	Visual aesthetic	photorealistic, anime, oil painting
Lighting	Light conditions	golden hour, neon, moonlight
Camera	Shot/angle/lens	close-up, wide shot, bokeh
Details	Quality modifiers	highly detailed, 8k, masterpiece
Mood	Emotional atmosphere	serene, ominous, epic

Intensity Levels

Level	Gap Fill Rate	Description
minimal	25%	Add only critical missing elements
light	50%	Subtle enhancement
moderate	75%	Balanced enhancement (default)
full	100%	Fill all detected gaps

Anti-Pairing Examples

ZeroENH prevents semantically incoherent combinations:

• underwater will not get golden hour or sunset lighting
• cave will not get blue hour or sunset backlight
• pixel art will not get ray tracing or photogrammetry
• graveyard will not get joyful or playful mood

Usage Examples

Minimal Input

Input:  "a cat"
Seed:   42
Intensity: moderate

Output: "photorealistic a cat resting in a cozy room, soft window light,
        close-up portrait, highly detailed, peaceful atmosphere"

Partial Input (preserves existing elements)

Input:  "cyberpunk samurai, neon lighting, cinematic"
Seed:   12345
Intensity: moderate

Output: "cyberpunk samurai standing in a neon-lit alley, cinematic,
        neon lighting, low angle shot, 8k, intense atmosphere"

With LoRA Triggers (preserved unchanged)

Input:  "a woman ohwx_person XYZ"
Seed:   42
Intensity: moderate

Output: "a woman in a dark forest, photorealistic, golden hour lighting,
        medium shot, highly detailed, mysterious atmosphere, ohwx_person, XYZ"

Comparison with LLM Enhancement

Aspect	LLM Enhancement	ZeroENH
Latency	30-60 seconds	<100ms
Determinism	Non-deterministic	Perfect reproducibility
Cost	API call per prompt	Zero (local compute)
Auditability	Black box	Full visibility
Customization	Prompt engineering	JSON vocabulary (V2)
Offline	Requires API	Fully offline

Roadmap

• V1 (Complete): Hardcoded default vocabulary - proves determinism works
• V2 (Complete): Custom JSON profiles with theme-specific vocabularies
• Future: Additional themed profiles (fantasy.json, portrait.json, anime.json, etc.)

Related Projects

This node is part of the DJZ-Nodes ecosystem and shares the position-is-seed methodology with:

• DJZ-ZeroPrompt - Full prompt generation from nothing
• Added-Typos - Deterministic text corruption

License

MIT License - see LICENSE for details.

Credits

Developed by MushroomFleet

Based on the ZeroBytes Position-is-Seed methodology for deterministic procedural generation.

Citation

Academic Citation

If you use this codebase in your research or project, please cite:

@software{comfyui_zeroenh,
  title = {ComfyUI-ZeroENH: Deterministic Prompt Enhancement via Procedural Augmentation},
  author = {Drift Johnson},
  year = {2025},
  url = {https://github.com/MushroomFleet/ComfyUI-ZeroENH},
  version = {1.0.0}
}

Donate:

---

Support This Project

If you found this useful, please star the repo — it helps others discover it!