ComfyUI OneReward Node

A custom node for ComfyUI that integrates OneReward model for high-quality image inpainting, outpainting, and object removal.

✨ Features

• 🎨 Intelligent Image Inpainting: Powered by OneReward model based on FLUX architecture
• 🖼️ Image Restoration: Smart fill of blank areas with prompt-guided content
• 🧽 Object Removal: Automatically remove unwanted elements from images
• 🔄 CFG Support: Built-in Classifier-Free Guidance for enhanced generation quality
• ⚙️ Memory Optimization: Quantization and CPU offload support for consumer-grade GPUs (~16GB VRAM)
• 🚀 Flexible Control: True CFG parameter control for fine-tuning results

🔧 Node List

Core Nodes

• RH_OneReward_Loader: Load and initialize OneReward models with optimization options
• RH_OneReward_Sampler: Prompt-guided image inpainting and restoration generator
• RH_OneReward_Eraser: Intelligent object removal tool

🚀 Quick Installation

Step 1: Install the Node

# Navigate to ComfyUI custom_nodes directory
cd ComfyUI/custom_nodes

# Clone the repository
git clone https://github.com/HM-RunningHub/ComfyUI_RH_OneReward

# Install dependencies
cd ComfyUI_RH_OneReward
pip install -r requirements.txt

Step 2: Download Required Models

# Download FLUX.1-Fill-dev model (Required base model)
huggingface-cli download black-forest-labs/FLUX.1-Fill-dev --local-dir models/black-forest-labs/FLUX.1-Fill-dev

# Download OneReward models
# Download standard OneReward model
huggingface-cli download black-forest-labs/flux.1-fill-dev-OneReward-transformer --local-dir models/OneReward/flux.1-fill-dev-OneReward-transformer

# Download dynamic OneReward model (optional)
huggingface-cli download black-forest-labs/flux.1-fill-dev-OneRewardDynamic-transformer --local-dir models/OneReward/flux.1-fill-dev-OneRewardDynamic-transformer

# Final model structure should look like:
models/
├── black-forest-labs/
│   └── FLUX.1-Fill-dev/
│       ├── text_encoder/
│       ├── text_encoder_2/
│       ├── tokenizer/
│       ├── tokenizer_2/
│       ├── transformer/
│       ├── vae/
│       └── scheduler/
└── OneReward/
    ├── flux.1-fill-dev-OneReward-transformer/
    └── flux.1-fill-dev-OneRewardDynamic-transformer/
    
# Restart ComfyUI

📖 Usage

Basic Workflow

[RH_OneReward_Loader] → [RH_OneReward_Sampler] → [Save Image]

Generation Types

Image Inpainting

• Load image that needs filling
• Provide mask layer marking areas to be filled
• Input text prompt describing the fill content
• Generate intelligently filled images

Image Restoration

• Load damaged or incomplete images
• Use mask to mark areas needing repair
• Input prompt describing restoration content
• Generate naturally restored images

Object Removal

• Use RH_OneReward_Eraser node
• Load image containing unwanted elements
• Provide mask marking areas to remove
• Automatically remove and fill background

🛠️ Technical Requirements

• GPU: 16GB+ VRAM (with quantization optimization)
• RAM: 32GB+ recommended
• Storage: ~30GB for all models
  • FLUX.1-Fill-dev: ~24GB
  • OneReward models: ~6GB
• CUDA: Required for optimal performance

⚠️ Important Notes

• Model Paths: Models must be placed in specific directories:
  • FLUX.1-Fill-dev → models/black-forest-labs/FLUX.1-Fill-dev/
  • OneReward models → models/OneReward/
• Low-memory GPUs automatically enable quantization and CPU offload optimization
• All model files must be downloaded before first use
• Supports true_cfg parameter for quality control

📄 License

This project is licensed under Apache 2.0 License.

🔗 References

• OneReward Project Page
• FLUX.1-Fill Model
• OneReward HuggingFace
• ComfyUI

🤝 Contributing

Contributions are welcome! Please feel free to submit issues and pull requests.

⭐ Citation

If you find this project useful, please consider citing the related papers:

@article{onereward2024,
    title={OneReward: Unified Image Inpainting and Outpainting},
    author={OneReward Team},
    year={2024},
    journal={arXiv preprint},
}