Real-ESRGAN is the hugely popular free, open-source super-resolution project from Tencent ARC Lab — with over 37,000 GitHub stars, it powers countless upscaling tools and restores real-world photos and videos.
Real-ESRGAN extends the ESRGAN architecture into a practical restoration system trained entirely on synthetic degradations, so it handles the messy reality of old, compressed, noisy images — not just clean test data. The flagship RealESRGAN_x4plus model enlarges images 4x while removing noise and JPEG artifacts, and the project also ships a 2x model, a compact anime model (RealESRGAN_x4plus_anime_6B) and AnimeVideo-v3 models for animation footage.
It integrates with GFPGAN for optional face enhancement, supports arbitrary output scales, and can be fine-tuned on your own data. Many well-known upscalers — including the GUI app Upscayl — are built on Real-ESRGAN’s models, which makes this repository the source much of the ecosystem flows from.
The repository remains active, with newer additions such as the compact realesr-general-x4v3 model, which adds adjustable denoising strength for general scenes. Its research paper, from Tencent ARC Lab and the Shenzhen Institutes of Advanced Technology, is still a standard citation in the super-resolution field years after publication.
Real-ESRGAN is completely free and open source. You can run it locally via Python, use the portable executables, or try hosted demos on Replicate, Hugging Face Spaces and Google Colab.
Real-ESRGAN is for developers, researchers and power users who want state-of-the-art upscaling they can run, inspect and build on — and for anyone curious where their favorite upscaler’s magic actually comes from.