Skip to main content
Real-ESRGAN Logo

​
Real-ESRGAN

Real-ESRGAN aims at developing Practical Algorithms for General Image/Video Restoration. We extend the powerful ESRGAN to a practical restoration application, which is trained with pure synthetic data. Real-ESRGAN Comparison

​
Key Features

Multiple Models

Specialized models for general images, anime illustrations, and anime videos

Easy Installation

Simple pip installation with Python >= 3.7 and PyTorch >= 1.7

Quick Start

Get started with upscaling in just a few lines of code

Face Enhancement

Integrated GFPGAN support for enhancing faces in images

​
Available Models

Real-ESRGAN provides several pre-trained models optimized for different use cases:

RealESRGAN_x4plus

The default model for general image super-resolution with 4x upscaling. Best for photographs and natural images.
Optimized for anime illustrations with a smaller model size (6 blocks). Provides excellent results for anime-style artwork.
Specialized model for anime videos with a compact architecture. Ideal for processing anime video frames.
A tiny model for general scenes with denoising support. Use the -dn option to balance noise and avoid over-smooth results.
2x upscaling model for cases where 4x is too aggressive.
Alternative 4x model with different training characteristics.

​
Research Paper

Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic DataAuthors: Xintao Wang, Liangbin Xie, Chao Dong, Ying ShanAffiliation: Tencent ARC Lab; Shenzhen Institutes of Advanced Technology, Chinese Academy of SciencesPublication: International Conference on Computer Vision Workshops (ICCVW), 2021
Read the Paper | Watch YouTube Video | View Poster

​
Try It Online

Before installing locally, you can try Real-ESRGAN online:

​
Next Steps

Installation

Install Real-ESRGAN and its dependencies

Quick Start

Run your first image upscaling example

​
Citation

If you use Real-ESRGAN in your research, please cite: 
@InProceedings{wang2021realesrgan,
    author    = {Xintao Wang and Liangbin Xie and Chao Dong and Ying Shan},
    title     = {Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data},
    booktitle = {International Conference on Computer Vision Workshops (ICCVW)},
    date      = {2021}
}

Build docs developers (and LLMs) love

Get started for freeTalk to us