Extensions/DJZ-ENH-system
ComfyUI Extension

DJZ-ENH-system

will be added to DJZ-Nodes, provided separately here

By MushroomFleet·Created about a year ago·Updated 5 months ago· 2
MushroomFleet/DJZ-ENH-system
Nodes2
On cloudLocal install
CategoryCustom-Nodes, DJZ-Nodes
Stars2
Updated5 months ago

Nodes (2)

ENH JSON Prompt Extractor
Custom-Nodes
Zenkai-ENH
DJZ-Nodes
Readme

DJZ_ENH - Enhanced Prompt Management System for ComfyUI

A powerful ComfyUI custom node package for managing and utilizing advanced system prompts with multi-shot enhancement capabilities, designed specifically for Vision Language Models (VLMs) and dynamic prompt workflows.

Features

  • Dynamic System Prompt Loading: Browse and load system prompts from organized markdown files
  • Multi-Prompt JSON Extraction: Parse and extract multiple enhanced prompts from structured JSON responses
  • Vision Language Model Integration: Optimized templates for VLMs like Qwen2.5VL
  • LoRA-Aware Prompting: Enhanced accuracy through trained caption knowledge integration
  • Template-Based Enhancement: Consistent multi-angle prompt generation with cinematic perspectives

Installation

  1. Navigate to your ComfyUI custom nodes directory:

    cd ComfyUI/custom_nodes/

  2. Clone or download this repository:

    git clone https://github.com/MushroomFleet/DJZ-ENH-system

  3. Restart ComfyUI

Core Components

1. ZenkaiENH Node (ZenkaiENH.py)

Purpose: Dynamically loads system prompts from the ENH-store/ directory

Features:

  • Automatic detection of .md files in ENH-store folder
  • Dropdown selection interface for easy prompt switching
  • Real-time file change detection
  • UTF-8 encoding support
  • Automatic folder creation

Inputs:

  • enh_file: Dropdown list of available .md prompt files

Outputs:

  • STRING: Complete content of the selected prompt file

Usage:

  1. Place your system prompt .md files in the ENH-store/ folder
  2. Connect the ZenkaiENH node to your Ollama/API nodes
  3. Select desired prompt from dropdown
  4. System prompt is automatically loaded and sent to output

2. ENH_JSON Node (ENH_JSON.py)

Purpose: Extracts multiple prompts from JSON-formatted LLM responses

Features:

  • Robust JSON parsing with markdown code block handling
  • Multiple format support (direct keys, array format, object format)
  • Automatic fallback to safe default prompts
  • Enhanced error handling and logging
  • Trailing comma and quote cleanup

Inputs:

  • json_text: Multi-line string containing JSON response from LLM

Outputs:

  • output1: First extracted prompt (Standard Shot)
  • output2: Second extracted prompt (Closeup Shot)
  • output3: Third extracted prompt (Wide Shot)
  • output4: Fourth extracted prompt (Action Shot)

Supported JSON Formats:

// Format 1: Direct keys
{
  "prompt1": "description here",
  "prompt2": "description here",
  "prompt3": "description here",
  "prompt4": "description here"
}

// Format 2: Array format
{
  "prompts": ["desc1", "desc2", "desc3", "desc4"]
}

// Format 3: Object array format
{
  "prompts": [
    {"prompt": "desc1"},
    {"prompt": "desc2"},
    {"prompt": "desc3"},
    {"prompt": "desc4"}
  ]
}

3. ENH-store Templates

The ENH-store/ directory contains system prompt templates designed for specific enhancement workflows.

Example: t2i_ENH-vision-4shot.md

Purpose: Multi-angle image prompt enhancement for VLMs

Key Features:

  • Vision analysis protocol for image inputs
  • Four distinct camera perspectives (Standard, Closeup, Wide, Action)
  • Consistency rules across all generated prompts
  • Professional photography and cinematography terminology
  • 80-100 word target length per prompt

Shot Types Generated:

  1. Standard Shot: Medium shot, balanced composition
  2. Closeup Shot: Portrait perspective, detailed features
  3. Wide Shot: Environmental context, establishing shot
  4. Action Shot: Dynamic movement, dramatic angles

Workflow Integration

Basic Workflow Setup

  1. Load System Prompt:

    • Add ZenkaiENH node
    • Select desired enhancement template
    • Connect output to your LLM node (Ollama/API)

  2. Process User Input:

    • Send user prompt + optional image to LLM
    • LLM processes using loaded system prompt
    • LLM returns JSON with 4 enhanced prompts

  3. Extract Multiple Prompts:

    • Add ENH_JSON node
    • Connect LLM JSON output to json_text input
    • Use the 4 outputs for different image generations

Advanced Multi-Shot Workflow

[User Input] → [ZenkaiENH] → [LLM Node] → [ENH_JSON] → [4x Image Generation]
                    ↓              ↓           ↓
              [System Prompt] [JSON Response] [4 Enhanced Prompts]

Benefits:

  • Consistent subject details across all variations
  • Professional cinematic perspectives
  • Enhanced composition and framing
  • Reduced prompt engineering time

Creating Custom Templates

Template Structure

  1. Header Section: Clear description of template purpose
  2. Input/Output Specification: Expected inputs and JSON format
  3. Processing Rules: Enhancement guidelines and consistency requirements
  4. Technical Specifications: Quality descriptors, style guidelines
  5. Example Output: Sample JSON structure

Template Guidelines

  • Use markdown format (.md files)
  • Place files in ENH-store/ directory
  • Include clear JSON output format specifications
  • Define consistency rules for multi-prompt generation
  • Specify target word counts and quality standards

LoRA Integration

For LoRA-aware prompting:

  1. Include LoRA caption knowledge in system prompt
  2. Reference trained descriptions for alignment
  3. Leverage vision analysis for consistency
  4. Maintain coherency across prompt variations

Best Practices

System Prompt Design

  • Keep JSON output rules clearly defined
  • Maintain flexibility for different input types
  • Include fallback behaviors for edge cases
  • Test with various VLM models

Workflow Optimization

  • Use consistent naming conventions for prompt files
  • Organize templates by use case (t2i, style, character, etc.)
  • Test multi-prompt outputs for consistency
  • Monitor LLM token usage

Quality Control

  • Verify JSON output format compliance
  • Check prompt consistency across all outputs
  • Validate enhancement quality improvements
  • Test with different image inputs

Troubleshooting

Common Issues

No .md files found:

  • Ensure files are placed in ENH-store/ directory
  • Check file extensions are .md
  • Restart ComfyUI after adding files

JSON parsing errors:

  • Check LLM output format matches expected JSON
  • Verify no markdown code blocks in JSON
  • Ensure proper JSON syntax (no trailing commas)

Inconsistent prompts:

  • Review system prompt consistency rules
  • Check if VLM is following template guidelines
  • Adjust temperature/generation parameters

Empty outputs:

  • Verify JSON contains required keys
  • Check fallback prompt functionality
  • Review LLM response completeness

Technical Requirements

  • ComfyUI: Latest version recommended
  • Python: 3.8+ with UTF-8 support
  • VLM Support: Qwen2.5VL, GPT-4V, or similar vision models
  • Memory: Sufficient for multi-prompt processing

Contributing

  1. Create new templates in ENH-store/ directory
  2. Follow existing template structure and naming conventions
  3. Test with multiple VLM models
  4. Document specific use cases and requirements

License

[Add your license information here]

Changelog

v1.0.0 - Initial Release

  • ZenkaiENH system prompt loader
  • ENH_JSON multi-prompt extractor
  • t2i_ENH-vision-4shot template
  • Basic workflow integration
  • LoRA-aware prompting foundation

This system transforms single prompts into professional, multi-perspective image generation workflows while maintaining consistency and quality across all outputs.

Support This Project

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

Star on GitHub