GFPGAN
Practical blind face restoration using a generative facial prior.
About
GFPGAN (GFP-GAN: Towards Real-World Blind Face Restoration with Generative Facial Prior) is an open-source project from Tencent's Applied Research Center by Xintao Wang, Yu Li, Honglun Zhang, and Ying Shan. It restores degraded faces in real-world photos by leveraging a generative facial prior learned from high-quality face data.
It handles blind restoration of old, blurry, or low-quality portraits, with optional Real-ESRGAN background enhancement. A clean version ships without custom CUDA extensions, and it is widely used for photo restoration and fixing AI-generated faces.
What it can do
- Blind face restoration from degraded photos
- Generative facial prior for realistic detail
- Background enhancement via Real-ESRGAN
- No custom CUDA extensions needed
Setup
Clone the repo, pip install basicsr, facexlib, and requirements, then run inference.
Prerequisites
Python 3.7+, PyTorch 1.7+, and optionally an NVIDIA GPU.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 103; Trending score 76/100 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.