EditMGT

🌟 Overview

This is the official repository for EditMGT: Unleashing the Potential of Masked Generative Transformer in Image Editing ✨.

EditMGT is a novel framework that leverages Masked Generative Transformers for advanced image editing tasks. Our approach enables precise and controllable image modifications while preserving original content integrity.

✨ Features

• 🎨 Great style transfer capabilities
• 🔍 Attention control over editing regions
• ⚡ The model backbone is only 960M, resulting in fast inference speed.
• 📊 Trained on the CrispEdit-2M dataset

⚡ Quick Start

First, clone the repository and navigate to the project root:

git clone https://github.com/weichow23/editmgt
cd editmgt

🔧 Environment Setup

# Create and activate conda environment
conda create --name editmgt python=3.9.2
conda activate editmgt

# Optional: Install system dependencies
sudo apt-get install libgl1-mesa-glx libglib2.0-0 -y

# Install Python dependencies
pip3 install git+https://github.com/openai/CLIP
pip3 install -r requirements.txt

⚠️ Note: If you encounter any strange environment library errors, please refer to Issues to find the correct version that might fix the error.

🚀 Inference

Run the following script in the editmgt directory:

import os
import sys
sys.path.append("./")
from PIL import Image
from src.editmgt import init_edit_mgt
from src.v2_model import negative_prompt

if __name__ == "__main__":
    pipe = init_edit_mgt(device='cuda:0')
    # Forcing the use of bf16 can improve speed, but it will incur a performance penalty. 
    # We noticed that GEditBench dropped by about 0.8.
    # pipe = init_edit_mgt(device, enable_bf16=False)

    # pipe.local_guidance=0.01 # After starting, it will use the local GS auxiliary mode.
    # pipe.local_query_text = 'owl' # Use specific words as attention queries
    # pipe.attention_enable_blocks = [i for i in range(28, 37)]  # attention layer used
    input_image = Image.open('assets/case_5.jspg')
    result = pipe(
        prompt=['Make it into Ghibli style'],
        height=1024,
        width=1024,
        num_inference_steps=36, # For some simple tasks, 16 steps are enough!
        guidance_scale=6,
        reference_strength=1.1,
        reference_image=[input_image.resize((1024, 1024))],
        negative_prompt=negative_prompt or None,
    )
    output_dir = "./output"
    os.makedirs(output_dir, exist_ok=True)

    file_path = os.path.join(output_dir, f"edited_case_5.png")
    w, h = input_image.size
    result.images[0].resize((w, h)).save(file_path)

📑 Citation

@article{chow2025editmgt,
  title={EditMGT: Unleashing Potentials of Masked Generative Transformers in Image Editing},
  author={Chow, Wei and Li, Linfeng and Kong, Lingdong and Li, Zefeng and Xu, Qi and Song, Hang and Ye, Tian and Wang, Xian and Bai, Jinbin and Xu, Shilin and others},
  journal={arXiv preprint arXiv:2512.11715},
  year={2025}
}

🙏 Acknowledgements

We extend our sincere gratitude to all contributors and the research community for their valuable feedback and support in the development of this project.