GFPGAN — AI Face Restoration & Enhancement Model 🖼️🔢❓ → 🖼️

▶️ 115.6M runs 📅 Sep 2021 ⚙️ Cog 0.4.2 🔗 GitHub 📄 Paper ⚖️ License
face-detection face-restoration image-restoration image-upscaling face-enhancement portrait

Related Guides

Performance

1.7sTypical run time
115.6MTotal runs

Data sampled 2026-07-12

GFPGAN (Generative Facial Prior GAN) is one of the most widely used face restoration models. Developed by TencentARC, it uses pretrained face generation knowledge to fill in missing facial details that simpler upscaling methods cannot recover.

About

GFPGAN by TencentARC is a practical face restoration model that enhances and restores degraded face images. It leverages generative facial priors from a pretrained face GAN (StyleGAN2) to recover realistic facial details — eyes, mouth, hair — from low-quality, blurry, or compressed face photos.

Key capabilities

  • Face restoration — recover detail from blurry, noisy, or low-resolution face images
  • Face enhancement — improve clarity and sharpness of portrait photos
  • Old photo repair — restore faces in vintage or damaged photographs
  • Video frame enhancement — improve face quality in individual video frames

GFPGAN balances fidelity (staying close to the input) with quality (generating realistic details), making it suitable for both creative and restoration workflows.

Example Output

Output

Example output

Performance Metrics

1.67s Prediction Time
1.97s Total Time
All Input Parameters 
{
  "img": "https://replicate.delivery/mgxm/59d9390c-b415-47e0-a907-f81b0d9920f1/187400315-87a90ac9-d231-45d6-b377-38702bd1838f.jpg",
  "scale": 2,
  "version": "v1.4"
}
Input Parameters
img (required) Type: string
Input
scale Type: numberDefault: 2
Rescaling factor
version Default: v1.4
GFPGAN version. v1.3: better quality. v1.4: more details and better identity.
Output Schema

Output

Type: string • Format: uri

Example Execution Logs 
/tmp/tmpky22vsq_187400315-87a90ac9-d231-45d6-b377-38702bd1838f.jpg v1.4 2.0
Version Details
Version ID
0fbacf7afc6c144e5be9767cff80f25aff23e52b0708f17e20f9879b2f21516c
Version Created
March 14, 2024
Run on Replicate →