Image
nextgenUp
Free local upscaler — 8K photos and 4K video with Real-ESRGAN, no cloud
★ 65 starsMIT
View guide →Real-ESRGAN — practical blind super-resolution for real-world photos and video: restores and upscales degraded images with models trained on pure synthetic data, BSD-3-Clause
Real-ESRGAN is an open-source image project by xinntao (xinntao/Real-ESRGAN on GitHub). It has 37,147 GitHub stars and 4,495 forks, is released under the BSD-3-Clause license, and was last updated on GitHub on 6 August 2024.
Clone the repo and install the requirements, or install the realesrgan package from PyPI (with basicsr, facexlib and gfpgan) per the README dependencies section.
Fetch the pretrained weights for your job — RealESRGAN_x4plus for general photos, the anime_6B model for illustrations, or the general-x4v3 model when you want denoising control with -dn.
Run inference_realesrgan.py with your input folder and chosen model; use --outscale for an arbitrary final scale and --face_enhance to chain GFPGAN face restoration.
Trained on synthetic degradations that mimic real damage, so it restores photos — not just clean downscales — convincingly.
General x4plus, a compact anime model, anime-video models and the denoise-tunable realesr-general-x4v3, all in the model zoo.
Integrates GFPGAN so portraits get face-aware restoration in the same pass with --face_enhance.
Official ncnn-vulkan executables for Windows, Linux and macOS run upscaling without any Python environment.
The package route from the README ecosystem.
pip install basicsr facexlib gfpgan
pip install realesrganThe repo route: install requirements, then run the inference script on a folder of images.
git clone https://github.com/xinntao/Real-ESRGAN.git
cd Real-ESRGAN
pip install -r requirements.txt
python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs -o resultsDownload realesrgan-ncnn-vulkan for your OS from the v0.2.5.0 release and run it directly — no Python needed.
https://github.com/xinntao/Real-ESRGAN/releasesXintao Wang and co-authors at Tencent ARC Lab and the Shenzhen Institutes of Advanced Technology (xinntao/Real-ESRGAN on GitHub); the paper is arXiv 2107.10833.
Yes — BSD-3-Clause licensed, including for commercial use, subject to the licence terms. Check the licences of any model weights you add.
Honestly: it is a mature, stable codebase — its last push was 2024-08-06 — rather than an actively evolving one. It remains the reference implementation and is embedded in many other tools.
Against GFPGAN from the same lab family, Real-ESRGAN is the general image/video restorer and GFPGAN is the face specialist — they are designed to chain together. Against newer diffusion upscalers, Real-ESRGAN is far faster and more predictable, with less invented detail.
pip install realesrgan (plus basicsr, facexlib, gfpgan), download the x4plus weights, and run inference_realesrgan.py on a test folder — or grab the portable ncnn build if you want zero setup.
Free local upscaler — 8K photos and 4K video with Real-ESRGAN, no cloud
Real-ESRGAN image and video upscaler app with GPU acceleration
Simple Flask web app that upscales low-res images with AI
DeepSeek's Janus series - unified multimodal models that both understand and generate images
🍌 World's largest Nano Banana Pro prompt library — 10,000+ curated prompts with preview images, 16 languages. Google Gemini AI image generation. Free & open …
A general fine-tuning kit geared toward image/video/audio diffusion models.
The most-starred other Image projects in the LibreAI catalogue — the realistic shortlist if Real-ESRGAN is not the right fit.
Stable Diffusion web UI
The node-graph powerhouse for precise, professional AI image and video generation.
High performance self-hosted photo and video management solution
Focus on prompting and generating
🆙 Upscayl - #1 Free and Open Source AI Image Upscaler for Linux, MacOS and Windows
Free and open — grab the source, star the repo, and start creating.