Check out open-source AI image upscaler Real-ESRGAN
Researchers at China’s Tencent ARC Lab, Shenzhen Institute of Advanced Technology and Chinese Academy of Sciences have released Real-ESRGAN, an interesting open-source AI image upscaling tool.
The software, which runs on Windows, Linux and macOS, and on AMD, Intel and Nvidia GPUs, comes with dedicated AI models for upscaling and removing noise from photos, 2D illustrations and 2D animation.
Real-ESRGAN itself has been available since 2021, but it has recently been attracting interest in the CG community, in part thanks to new frontends including an integration with free image viewer XnView.
Upscale 16-bit images, including photos, 2D illustrations and anime
Released alongside the research paper on which it is based, Real-ESRGAN is intended to restore degraded images, such as low-resolution copies (or copies of copies) found online.
It’s system- and hardware-agnostic, and supports 16-bit images, including images with alpha channels.
It comes with a range of AI models, including models specialised on photos – it integrates Tencent ARC’s GFPGAN for face restoration – 2D illustrations and 2D animation, specifically anime.
Use Real-ESRGAN from within XnView with Real ESRGAN Bat
Although, unlike commercial AI upscaling tools like Gigapixel AI, there is no dedicated AI model for upscaling rendered 3D images, Real-ESRGAN has been attracting attention in the CG community recently.
Above, you can see a walkthrough on installing and using the software from YouTube news channel askNK.
It can also be called from inside XnView, making it possible to upscale images inside the free image viewer.
By default, it upscales an image 4x using the AI model for 2D illustrations, but you can change that by editing the parameters and file path in the .bat file.
System requirements and availability
Real-ESRGAN 0.2.5 is available under BSD 3-Clause licence. You can download compiled executables for Windows, Linux and macOS from GitHub. Real ESRGAN Bat is a free download.