GFPGAN: The AI Face Restoration Tool With 37k Stars That Actually Preserves Identity

What Is GFPGAN?

Have an old family photo with faces so blurry you can barely recognize them? GFPGAN can fix that. With over 37,000 GitHub stars, it's the most popular open-source AI face restoration tool — and it produces results that genuinely look like the original person, not a smoothed-out mannequin.

Developed by TencentARC, GFPGAN uses a clever combination of StyleGAN (a face generation model) and a U-Net restoring network. The StyleGAN provides "prior knowledge" about what human faces should look like — eye shapes, nose structures, skin textures — while the U-Net handles the noisy, degraded input. The result is face restoration that preserves identity while adding realistic detail that was lost to compression, blur, or low resolution.

The teaser image above shows what GFPGAN can do. On the left, blurry, compressed faces. On the right, the same faces after restoration — note how facial features are sharpened, skin texture is realistic, and identity is preserved.

Why GFPGAN Is the Best Face Restoration Tool

1. StyleGAN Face Prior

This is GFPGAN's secret weapon. The model uses a pre-trained StyleGAN as a "dictionary of faces" — it knows what realistic eyes, noses, mouths, and skin should look like. When restoring a degraded face, it pulls from this prior to add realistic detail rather than hallucinating artifacts. This is why GFPGAN results look like the actual person, not a generic smoothed face.

2. Rescues Severely Degraded Faces

GFPGAN handles extreme cases that simpler upscalers fail on. Faces at 32x32 pixels, heavy JPEG compression artifacts, motion blur — GFPGAN reconstructs recognizable facial features from inputs most tools would give up on. It's particularly good at restoring eyes and mouths, which are critical for facial recognition and emotional expression.

3. Identity Preservation

Unlike generic upscalers that produce "average" faces, GFPGAN actively works to preserve identity. The model includes a facial component loss that ensures the restored face maintains the same eye shape, nose structure, and facial geometry as the original. This is essential for restoring photos of real people where identity matters.

4. Video Face Restoration

GFPGAN can process video frames, restoring faces frame by frame while maintaining temporal consistency. This makes it useful for restoring old videos, enhancing surveillance footage, and cleaning up video calls. The GPU-accelerated processing handles real-time video at 720p on modern GPUs.

5. Works with Real-ESRGAN Integration

GFPGAN integrates directly with Real-ESRGAN for complete photo restoration. Real-ESRGAN handles the background, textures, and overall image upscaling, while GFPGAN focuses specifically on restoring faces within the image. Together, they produce fully restored photos with enhanced backgrounds and crisply detailed faces.

Getting Started

Installation

git clone https://github.com/TencentARC/GFPGAN.git
cd GFPGAN
pip install -r requirements.txt
python setup.py develop

Restore a Face Image

python inference_gfpgan.py -i inputs/ -o results/ -v 1.3 -s 2

• -i: input folder (put your blurry face photos here)
• -o: output folder (restored photos appear here)
• -v: version (1.3 = best quality, 1.2 = faster)
• -s: upscale factor (2 for 2x, 4 for 4x)

Python API

import cv2
from gfpgan import GFPGANer

restorer = GFPGANer(
    model_path='experiments/pretrained_models/GFPGANv1.3.pth',
    upscale=2, arch='clean', channel_multiplier=2,
    bg_upsampler=None  # use RealESRGANer here for full restoration
)

img = cv2.imread('old_photo.jpg', cv2.IMREAD_COLOR)
_, _, restored_img = restorer.enhance(img, has_aligned=False, only_center_face=False, paste_back=True)
cv2.imwrite('restored_photo.jpg', restored_img)

Self-Hosting with Docker

docker pull wpafbo79/gfpgan:latest

docker run -d \
  --name gfpgan \
  --gpus all \
  -v ./inputs:/inputs \
  -v ./outputs:/outputs \
  wpafbo79/gfpgan:latest

GFPGAN vs Other Face Restoration Tools

Best Use Cases

Old Photo Restoration

Bring blurry family photos from decades ago back to life. GFPGAN handles the faces, while Real-ESRGAN can handle the rest of the photo.

Video Enhancement

Restore faces in old VHS recordings, security footage, or low-quality video calls.

Forensic Applications

Enhance facial details in surveillance and evidence photos where identity is critical.

Historical Archives

Restore faces in historical photographs, paintings, and scanned documents.

Tips for Best Results

• Use -v 1.3 for the best quality model (GFPGANv1.3)
• Set -s 2 for 2x upscale — it's the sweet spot for face restoration
• Pair with Real-ESRGAN as bg_upsampler for complete photo restoration
• Use only_center_face=False when restoring group photos
• For severely damaged photos, pre-process with denoising before GFPGAN

Conclusion

GFPGAN is the definitive face restoration tool in the open-source world. Its StyleGAN prior lets it reconstruct facial details that other tools simply can't — and it preserves identity while doing so. With 37k GitHub stars, video support, and integration with Real-ESRGAN for full photo restoration, it's the tool of choice for anyone serious about fixing faces in photos and videos.