ComfyUI Extension: ComfyUI-PromptEnhancer

Authored by ryan-seungyong-lee

Created 6 days ago

Updated 6 days ago

1 stars

ComfyUI-PromptEnhancer is a custom node for Hunyuan PromptEnhancer

Custom Nodes (0)

README

ComfyUI-PromptEnhancer

ComfyUI custom nodes for Hunyuan PromptEnhancer - Enhance your text-to-image prompts and image editing instructions with AI-powered prompt rewriting.

🌟 Features

🎨 Text-to-Image Prompt Enhancement: Transform simple prompts into detailed, well-structured descriptions
🖼️ Image-to-Image Editing Enhancement: Enhance editing instructions while preserving non-edited regions
🤖 Automatic Model Download: Models are automatically downloaded from HuggingFace when first used
🌐 Multi-language Support: Works with both Chinese and English prompts

📦 Installation

⚠️ Important: Install flash-attn (Required for Image-to-Image)

Note: This is not required for Text-to-Image Prompt Enhancement (7B).

The Image-to-Image Prompt Enhancer node requires flash-attn. Install it first:

pip install flash-attn --no-build-isolation

Method 1: ComfyUI Manager (Recommended)

Install ComfyUI Manager
In ComfyUI, go to Manager → Install Custom Nodes
Search for "PromptEnhancer"
Click Install and restart ComfyUI

Method 2: Git Clone

cd ComfyUI/custom_nodes/
git clone https://github.com/ryan-seungyong-lee/ComfyUI-PromptEnhancer.git
cd ComfyUI-PromptEnhancer
pip install -r requirements.txt

Method 3: Manual Installation

Download this repository as ZIP
Extract to ComfyUI/custom_nodes/ComfyUI-PromptEnhancer/
Install dependencies:

cd ComfyUI/custom_nodes/ComfyUI-PromptEnhancer
pip install -r requirements.txt

🚀 Quick Start

Available Nodes

1. Prompt Enhancer (Text-to-Image)

Enhances text prompts for better image generation results.

Inputs:

prompt: Your original prompt (text)
model_name: Choose between "promptenhancer-7b" (default, faster) or "promptenhancer-32b" (higher quality)
temperature: Sampling temperature (0 = deterministic, higher = more creative)
top_p: Nucleus sampling parameter
max_new_tokens: Maximum length of enhanced prompt
system_prompt (optional): Custom system prompt for specialized use cases

Output:

enhanced_prompt: Enhanced, detailed prompt ready for image generation

2. Prompt Enhancer (Image-to-Image)

Enhances image editing instructions based on visual context.

Inputs:

image: Input image to edit
edit_instruction: Your editing instruction (text)
temperature: Sampling temperature
top_p: Nucleus sampling parameter
max_new_tokens: Maximum length of enhanced instruction
system_prompt (optional): Custom system prompt

Output:

enhanced_prompt: Enhanced editing instruction preserving non-edited regions

📥 Model Downloads

Models are automatically downloaded on first use and cached in ComfyUI/models/prompt_enhancer/.

| Model | Size | Quality | Download Time* | Best For | |-------|------|---------|---------------|----------| | promptenhancer-7b | ~13GB | High | ~15 min | Most users, balanced performance | | promptenhancer-32b | ~64GB | Highest | ~60 min | Maximum quality, requires 32GB+ VRAM | | promptenhancer-img2img | ~30GB | High | ~30 min | Image editing tasks |

*Approximate times on a 100Mbps connection

Manual Download (Optional)

If you prefer to download models manually:

# For Text-to-Image (7B model)
huggingface-cli download tencent/HunyuanImage-2.1 --include "reprompt/*" --local-dir ComfyUI/models/prompt_enhancer/promptenhancer-7b

# For Text-to-Image (32B model)
huggingface-cli download PromptEnhancer/PromptEnhancer-32B --local-dir ComfyUI/models/prompt_enhancer/promptenhancer-32b

# For Image-to-Image
huggingface-cli download PromptEnhancer/PromptEnhancer-Img2img-Edit --local-dir ComfyUI/models/prompt_enhancer/promptenhancer-img2img

🎯 Usage Examples

Example Workflow

Check out the example workflow to get started quickly:

How to use:

Download the workflow image above
Open ComfyUI
Drag and drop the image into the ComfyUI canvas
The workflow will be automatically loaded with all nodes configured

Example 1: Basic Text Prompt Enhancement

Input:

"a cat"

Enhanced Output:

"A photorealistic portrait of a domestic cat with soft fur, sitting elegantly on a wooden surface. 
The cat has bright, alert eyes and pointed ears, with whiskers clearly visible. The composition 
focuses on the feline subject with natural lighting creating subtle shadows. The background is 
gently blurred to emphasize the cat. High-quality digital photography style."

Example 2: Gaming Scene Enhancement

Input (Chinese):

"第三人称视角，赛车在城市赛道上飞驰"

Enhanced Output:

"第三人称视角游戏画面，一辆流线型赛车正在繁华的城市赛道上高速飞驰。左上角显示小地图，
标注当前位置和赛道布局，地图下方实时显示当前排名。右下角的仪表盘清晰展示速度表，
指针指向高速区域。赛道两旁是现代化的城市建筑群，具有动态模糊效果增强速度感。
画面采用动态追踪镜头，呈现竞速游戏的紧张氛围和高画质视觉效果。"

Example 3: Image Editing Enhancement

Input Image: [Image with watermark] Instruction:

"去除图片中的水印"

Enhanced Output:

"保持画面中的主体内容、构图和色调完全不变，仅将图片底部的文字水印或logo标识完整移除。
使用智能修复技术填补去除水印后的区域，确保过渡自然无痕迹。其余所有元素包括前景、
背景、光影效果、人物或物体的位置、大小、颜色均保持原样。"

⚙️ Advanced Configuration

Custom System Prompts

You can provide custom system prompts for specialized use cases:

For Anime Style:

system_prompt = """你是一位专注于动漫风格图像提示词的专家。
请将用户输入改写为适合动漫风格图像生成的提示词，
注重角色特征、动漫画风特点和典型的视觉元素。"""

For Photography:

system_prompt = """You are an expert in photography prompt writing.
Rewrite the user's input into a detailed photography prompt,
including camera settings, lighting, composition, and style."""

Parameter Tuning Guide

temperature = 0: Deterministic output (recommended for consistency)
temperature = 0.7: Balanced creativity and consistency (recommended default)
temperature = 1.0+: More creative variations (may lose some intent)
top_p = 0.9: Standard nucleus sampling (recommended)
top_p = 1.0: Consider all tokens (more diverse but less focused)
max_new_tokens:
- 256-512: Short, concise enhancements
- 512-1024: Standard detailed descriptions (recommended)
- 1024-2048: Very detailed, structured prompts

🔧 Troubleshooting

Issue: Model download fails

Solution:

Check your internet connection
Check HuggingFace Authentication (hf auth login or huggingface-cli login)
Ensure you have enough disk space (7B model needs ~15GB free)
Try manual download using huggingface-cli (see Manual Download section)

Issue: Out of memory error

Solution:

Use the 7B model instead of 32B
Close other applications to free up VRAM
If using CPU, reduce max_new_tokens parameter

Issue: Node doesn't appear in ComfyUI

Solution:

Ensure all dependencies are installed: pip install -r requirements.txt
Restart ComfyUI completely
Check the console for error messages

Issue: Slow inference

Solution:

First run is slower due to model loading (subsequent runs use cached model)
Use the 7B model for faster inference
Reduce max_new_tokens if you don't need very long prompts
Ensure you have adequate VRAM (recommended 8GB+ for 7B model)

📊 System Requirements

Minimum Requirements (7B Model)

GPU: 8GB VRAM (NVIDIA GPU with CUDA support)
RAM: 16GB system RAM
Storage: 20GB free space
OS: Windows 10/11, Linux, macOS (with Metal support)

Recommended Requirements (7B Model)

GPU: 12GB+ VRAM (RTX 3060 12GB or better)
RAM: 32GB system RAM
Storage: SSD with 30GB+ free space

For 32B Model

GPU: 32GB+ VRAM (A100, H100, or multiple GPUs)
RAM: 64GB+ system RAM
Storage: 80GB+ free space

📚 Additional Resources

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is licensed under the Tencent Hunyuan Community License Agreement - see the LICENSE file for details.

🙏 Acknowledgements

Hunyuan-PromptEnhancer Team for the original implementation
ComfyUI for the amazing node-based interface
HuggingFace for model hosting

📞 Support

Issues: GitHub Issues
Discussions: GitHub Discussions
Original Project: [email protected]

⭐ Star History

Made with ❤️ for the ComfyUI community