ComfyUI MiniMax-Remover

A ComfyUI custom node for fast and effective video/image object removal using MiniMax optimization. Remove objects from videos and images with high quality results using only 6-12 inference steps.

✨ Features

🚀 Fast: Only 6-12 inference steps required, no CFG needed

🎯 Effective: Seamlessly removes objects from videos and images

💪 Robust: Prevents regeneration of unwanted objects or artifacts

🔧 Easy to Use: Simple ComfyUI nodes with intuitive workflows

📱 Flexible: Supports both single images and video processing

⚡ Optimized: Automatic model downloading and caching

💡 Smart Hints: Automatic backend suggestions for optimal settings

🛠️ Installation

Method 1: ComfyUI Manager (Recommended)

Open ComfyUI Manager

Search for “MiniMax-Remover”

Click Install

Restart ComfyUI

Method 2: Manual Installation

Navigate to your ComfyUI custom_nodes directory:

cd ComfyUI/custom_nodes

Clone this repository:

git clone https://github.com/1038lab/ComfyUI-MiniMax-Remover.git

Install dependencies:

cd ComfyUI-MiniMax-Remover

For Portable ComfyUI (Windows):

..\..\..\python_embeded\python.exe -m pip install -r requirements.txt

For Standard Python Installation:

pip install -r requirements.txt

Restart ComfyUI

📥 Model Download

Models will be automatically downloaded from HuggingFace when first used.

Manual Model Download (Optional)

If automatic download fails or you prefer manual installation:

Download using HuggingFace CLI:

huggingface-cli download zibojia/minimax-remover --include vae transformer scheduler --local-dir ./models/MiniMax-Remover

Or download manually from HuggingFace:

Visit: https://huggingface.co/zibojia/minimax-remover

Download the vae, transformer, and scheduler folders

Place them in: ComfyUI/models/MiniMax-Remover/

Verify installation:

Check that ComfyUI/models/MiniMax-Remover/ contains:

vae/ folder

transformer/ folder

scheduler/ folder

📦 Available Nodes

| Node Name | Display Name | Purpose |

|———–|————–|———|

| MinimaxImageRemover | MiniMax Image Object Remover | Remove objects from single images (requires external model loader) |

| MinimaxVideoRemover | MiniMax Video Object Remover | Remove objects from videos (requires external model loader) |

| MinimaxModelLoader | MiniMax Model Loader | Load VAE, Transformer, and Scheduler for separate nodes |

| MinimaxVideoLoader | MiniMax Video Loader | Load video files for processing |

| ImageSizeAdjuster | MiniMax Image Size Adjuster | Adjust image dimensions for compatibility |

⚙️ Recommended Settings

MiniMax Model Loader (Advanced)

scheduler_type: FlowMatchEulerDiscreteScheduler (recommended) or UniPCMultistepScheduler

torch_dtype: float16 (GPU recommended) or float32 (CPU/compatibility)

device: auto (automatic detection)

🎨 Mask Requirements

Supported Formats

✅ Grayscale images (recommended)

✅ RGB images (automatically converted)

✅ PNG, JPG, TIFF and other common formats

Mask Guidelines

Black background, white foreground (standard mask format)

High contrast for better edge detection

Complete coverage of objects to remove

Clean edges without excessive noise

Avoid overly complex details in small areas

🚀 Performance Tips

GPU Memory Optimization

Use float16 precision (reduces memory by ~50%)

Keep image resolution reasonable (≤1024×1024 recommended)

Use fewer inference steps (6-12 usually sufficient)

Close other GPU applications to free VRAM

Processing Speed

FlowMatchEulerDiscreteScheduler is typically faster than UniPCMultistepScheduler

temporal_frames=3 provides best speed/quality balance

Use ImageSizeAdjuster node to optimize image dimensions for better performance

🔧 Requirements

System Requirements

Python: 3.8 or higher

GPU: NVIDIA GPU with CUDA support (recommended)

RAM: 8GB+ system RAM

VRAM: 6GB+ GPU memory for optimal performance

Dependencies

All dependencies are automatically installed via requirements.txt:

torch>=1.13.0
diffusers>=0.21.0
decord>=0.6.0
einops>=0.6.0
scipy>=1.9.0
opencv-python>=4.5.0
huggingface_hub>=0.16.0
accelerate>=0.20.0

❓ Troubleshooting

Common Issues

Model Download Fails

Try manual download from HuggingFace website

Check internet connection and HuggingFace access

Ensure sufficient disk space in ComfyUI/models/MiniMax-Remover/

CUDA Out of Memory

Reduce image resolution

Use float16 instead of float32

Reduce num_inference_steps

Close other applications using GPU

Tensor Dimension Errors

Ensure mask and image have compatible dimensions

Enable fix_dimensions in the All-in-One node for automatic adjustment

Use ImageSizeAdjuster node for manual dimension control

Check that images are RGB (not RGBA)

Dimension Warning Messages

If you see “Image dimensions not divisible by 16” warning, enable fix_dimensions=True

The All-in-One node will automatically suggest this when needed

Poor Removal Quality

Increase num_inference_steps (try 16-20)

Improve mask quality (higher contrast, cleaner edges)

Adjust mask_dilation_iterations based on mask precision

Try different scheduler types

📚 Additional Resources

Original MiniMax-Remover Paper

HuggingFace Model Repository

ComfyUI Documentation

📄 License

This project is licensed under the GNU General Public License v3.0 – see the LICENSE file for details.

🙏 Acknowledgments

MiniMax-Remover Team for the original research and implementation

ComfyUI Community for the excellent framework and ecosystem

HuggingFace for model hosting and distribution

Contributors who helped improve this custom node

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

⭐ Support

If you find this project helpful, please consider giving it a star on GitHub! It helps others discover the project and motivates continued development.

---

Made with ❤️ for the ComfyUI community