Real-ESRGAN
Open-source blind super-resolution for photos and anime.
About
Real-ESRGAN is Tencent ARC Lab's open-source image and video restoration project, from the paper on training real-world blind super-resolution with pure synthetic data. It became the de facto standard engine for AI upscaling: general x4 models for photos, anime-tuned variants, face enhancement via GFPGAN integration, and an ncnn Vulkan build that runs without a heavy PyTorch stack.
The project underpins many consumer upscalers, including Upscayl, and ships training code so researchers can fine-tune on their own datasets. It is a research-grade tool rather than a polished product: setup means cloning the repo and installing PyTorch, but it remains the reference implementation most upscaling apps are measured against.
What it can do
- General and anime-tuned 4x super-resolution models
- Blind restoration of real-world degraded images
- Face enhancement through GFPGAN integration
- ncnn Vulkan builds for lighter deployment
- Training code for fine-tuning on custom data
Setup
Clone the GitHub repo and install PyTorch, or use the ncnn Vulkan build.
Prerequisites
Python and a CUDA-capable GPU for the full version; CPU/Vulkan builds available.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 90; Trending score 72/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.