Free AI tool makes old photos that have been altered by time!

Time to fix your grandparents photos

One of the most talked about applications of artificial intelligence algorithms is photos. For example, we recently informed you about AI that can convert 2D photos into 3D, offering almost surreal results.

Now, GFP-GAN (Generative Facial Prior-Generative Adversarial Network), a free and open source AI tool developed by researchers at China’s Tencent, is at the center of online debate. Said artificial intelligence can restore old, time-worn and low-resolution photos of people. Yes, it can practically bring back those old photos of your grandparents in your home.

Going into detail, the technology merges information from two artificial intelligence models to fill in the missing elements of a photo with realistic details, in just a few seconds, while maintaining high accuracy and quality. Conventional methods used until now focused on refining an existing artificial intelligence model for image restoration by comparing the differences between artificial and real photos. This often leads to low quality results according to Tencent scientists.

This new approach uses a pre-trained version of an existing model (NVIDIA’s StyleGAN-2) to update the team’s new model at multiple stages during the image generation process. The technique aims to preserve the “identity” of people, with particular emphasis on facial features such as the eyes and mouth.

Of course, it is worth noting that as impressive as the results are, this AI technology still has several limitations. Obviously, for the parts of the photos that are almost completely destroyed, he needs to make predictions that are based on the information that can be extracted from the surrounding areas, but the researchers warn that it is possible that the final result “slightly changes the identity” of the person who shown in the photo. Finally, another limitation is the resolution, which is not upgraded by the specific AI algorithm.

In case you are interested, you can try a demo of GFP-GAN completely free by clicking here . At the same time, since the project is open source, by clicking here you can download it, to use its technology in other ways.

Here’s a video demonstration that showcases the impressive capabilities of the algorithm.