Practical overview
When it fits
Use GFPGAN when the image task is specifically restoration of faces, including blind restoration of real-world inputs. The repository also documents optional Real-ESRGAN installation for background enhancement.
Who it is for
- Developers building face-restoration workflows for degraded portraits.
- Image-processing users who can install Python and PyTorch packages and run a command-line inference script.
- Practitioners who want face-focused enhancement and may optionally add background upsampling through Real-ESRGAN.
Documented uses
- Restore facial details in real-world images.
- Process one image or a folder of images through an inference script.
- Choose among listed model versions and set an output upsampling scale.
- Optionally enhance non-face background regions using Real-ESRGAN.
Requirements & limits
- The clean version does not require customized CUDA extensions; the original paper model has separate installation instructions.
- Python 3.7 or later and PyTorch 1.7 or later are listed prerequisites; Linux and an NVIDIA GPU with CUDA are optional in the installation notes.
- BasicSR and facexlib are installation dependencies; Real-ESRGAN is an additional dependency for background enhancement.
- The repository describes V1.3 as more natural and better on very low- and high-quality inputs, V1.2 as having no colorization and requiring no CUDA extensions, and V1 as the paper model with colorization.
- The README links hosted demonstrations as well as the local repository workflow.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run GFPGAN inference on images
Install the clean repository version, obtain a pretrained model, and run its documented inference command on an image or folder.
Before you start
- Python 3.7 or later
- PyTorch 1.7 or later
- BasicSR, facexlib and the repository requirements
- A pretrained GFPGAN model; the README example uses V1.3
- NVIDIA GPU with CUDA and Linux are optional
Steps
Clone the TencentARC/GFPGAN repository and change into its directory.
Install BasicSR, facexlib and the packages in requirements.txt, then run python setup.py develop.
Download GFPGANv1.3.pth into experiments/pretrained_models.
Run python inference_gfpgan.py -i inputs/whole_imgs -o results -v 1.3 -s 2, adjusting the input and output paths as needed.
Expected outcome
The inference script processes an image or folder and writes restored results to the selected output folder, using the chosen version and upsampling scale.
Guide limitations
- The documented command assumes the example input folder and a downloaded model; substitute paths for your own files.
- Background enhancement with Real-ESRGAN requires installing the realesrgan package.
- The README documents command-line options but does not guarantee a particular result for an input image.
Compare options for Model inference and serving
These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.
| Project | Documented purpose | Platform | Access and requirements |
|---|---|---|---|
| Gfpgan | Restore facial details in real-world images. | Linux | |
| FASHN VTON v1.5 | Generate a try-on preview from a person image and a garment image. | CLI, Linux, macOS, Self-hosted | The repository is public and states an Apache-2.0 license. Local setup requires downloading approximately 2 GB of model weights. |
| Posteromni | Rescale a poster to a target aspect ratio using a text prompt. |
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- TencentARC/GFPGAN · Official homepage title · retained in its source language