A free AI program that repairs old images by making somewhat fresh versions of your beloved
AI that generates new photographs is available, but what if you want to edit an old family photo? You may have a cost-free choice. GFP-GAN (Generative Facial Prior-Generative Adversarial Network), a recent free program created by Tencent researchers, may repair damaged and low-resolution photos, according to Louis Bouchard and PetaPixel. The system combines data from two AI models to quickly and accurately add realistic detail to a photo while preserving high accuracy and quality.
By comparing the disparities between the fabricated and actual images, conventional techniques refine an existing AI model to recover images. The scientists claimed that this typically resulted in subpar outcomes. A pre-trained version of an existing model (NVIDIA’s StyleGAN-2) is used in the new method to provide feedback to the team’s own model at various points along the picture generating process.
GFP-GAN has a free demo you may test out. To enable anybody to use the restoration technology in their own projects, the inventors have also made their source code available.
The constraints imposed by contemporary AI are still in place for this project. It makes educated assumptions about the missing material, but it’s remarkably accurate. In addition to a lower resolution than you may want, the researchers cautioned that you might notice a “slight alteration of identity.”
People, don’t count on this to print a grandparent’s photo the size of a billboard. However, the work presented here is encouraging because it raises the possibility of a time when it may be simple to save pictures that would otherwise be lost to the ravages of time.