Skip to content
RepoQuarry

GFPGAN TencentARC/GFPGAN

Tencent ARC face restoration model that repairs blurry, low-res faces in old photos, with optional Real-ESRGAN background upscaling.

Creator kit + Compare GitHub ↗ Install guide (official) ↗

GFPGAN at a glance

Checked 16 Oct 2026

GFPGAN restores faces in old or low-quality photos using a pretrained StyleGAN2 prior, and can upscale the rest of the image with Real-ESRGAN. It is a Python command-line research project (PyTorch, NVIDIA GPU optional); many GUIs and ComfyUI nodes reuse its weights. The main code is Apache-2.0, but its licence file lists StyleGAN2- and DFDNet-derived parts under non-commercial terms, and the weights have no separate licence. It can shift identity slightly. Mature but inactive: last release v1.3.8 (Sep 2022), last commit Jul 2024.

Cite this pageRepoQuarry. “GFPGAN.” Checked 16 Oct 2026. https://repoquarry.dev/projects/gfpgan

How much of Topaz Gigapixel does it replace?

2 of 6 jobs. Face recovery is its main job; background upscaling goes through Real-ESRGAN. Command line only.

Partial: Photo upscaling up to 16x Partial replacement
Partial: Face recovery on blurry or low-res faces Partial replacement
No: Recover and redefine low-quality or compressed images Not covered
No: Upscaling art Not covered
No: Batch upscaling of many images Not covered
No: Photoshop and Lightroom plugin Not covered

For recover and redefine low-quality or compressed images, upscaling art, batch upscaling of many images, Photoshop and Lightroom plugin, see other Topaz Gigapixel alternatives. You may also need to build an interface, an API server and hosting yourself.

Facts, with sources

Every answer links to where we found it. We read the documentation; we did not install it.

Report a correction
QuestionAnswerSourceChecked
Runs on my machine? Confirmed: Yes, Python script on your machine python inference_gfpgan.py -i inputs/whole_imgs -o results -v 1.3 -s 2 github.com · GFPGAN#zap-quick-inference ↗ today
Code licence Partial: Apache-2.0, except StyleGAN2- and DFDNet-derived parts under non-commercial licences GFPGAN is licensed under the Apache License Version 2.0 except for the third-party components listed below. github.com · LICENSE ↗ today
Model weights licence Unknown: The repository declares Apache-2.0 with third-party exceptions, but a separate blanket grant for all released checkpoints was not verified; original StyleGAN2-related components have additional restrictions GFPGAN is licensed under the Apache License Version 2.0 except for the third-party components listed below. github.com · LICENSE ↗ today
Commercial use Partial: Unclear; bundled third-party code includes NVIDIA and DFDNet non-commercial terms The Work and any derivative works thereof only may be used or intended for use non-commercially. github.com · LICENSE ↗ today
Paid services needed? Confirmed: None required github.com · GFPGAN#readme ↗ today
Hardware Confirmed: PyTorch; NVIDIA GPU with CUDA optional; the clean version needs no custom CUDA extensions Option: NVIDIA GPU + CUDA github.com · GFPGAN#wrench-dependencies-and-installation ↗ today
Operating systems Partial: Linux listed; Windows and macOS not documented Option: Linux github.com · GFPGAN#wrench-dependencies-and-installation ↗ today
Setup effort Confirmed: Medium, clone repo, pip install basicsr, facexlib and requirements, download weights We now provide a clean version of GFPGAN, which does not require customized CUDA extensions. github.com · GFPGAN#installation ↗ today

Get the weekly top 5 and what’s new, every Monday

We’ll email you once a week. Unsubscribe any time. We never share your email.

Cookie preferences

Choose what we may store in your browser. You can change this any time from Cookie settings in the footer.

Essential

Security, and remembering this choice. rq-consent · 12 months

Always on