GFP-GAN (Generative Facial Prior-Generative Adversarial Network), a new AI-based online tool developed by Tencent researchers, can help restore damaged and low-resolution portraits.
This new, freeware application can analyse the source image and fill in missing areas by adding pixels using a pre-configured generative model that takes the original back to the latent code and then reconstructs it. Additional processing enhances facial details and smoothes skin texture as well as improving the resolution of the person’s eyes, mouth and nose. The final step is to compare the source image to the newly restored image to check they are the same person.
While it can improve details, the developers admit the restored images may show a ‘slight change of identity’. This is more likely to occur with low-resolution originals and those that are severely damaged since the processing system has less data to work with.
The developers have made the source code for the application available to other developers here. You can also try out the system yourself here.
