"AI tool that restores all your old photos in seconds

 

You probably have a tonne of old photographs of your parents and grandparents if you're anywhere near my age (the photos, not your ancestors). Anyhow, without any prior Photoshop experience, you can now take those old, scratched, crinkled, and shredded prints and repair them.

Your outdated and damaged images can be accurately and quickly restored using a neural network known as the GFP-GAN (Generative Facial Prior-Generative Adversarial Network). However, this AI-based tool has drawbacks just like all others. Let's examine the advantages and disadvantages to discover what it has to offer.

In a blog post he made on the tool, Louis Bouchard called our attention to it. Of course, you can read more about this concept in depth in a research article as well. Simply said, when you upload your photo to GFP-GAN, it only generates a guess of the person's face. They appear to be rather close to the original image, nevertheless, in the vast majority of instances.

In order to add pixels or fill in the gaps, the AI first tries to grasp what is in the picture. GFP-GAN, in contrast to previous comparable models, emphasises crucial facial features like the eyes and mouth. The next step is to compare the final image to the original to determine if the subject is still the same.

Different technologies are used in conventional picture restoration techniques to reconstruct damaged or fuzzy photos and produce fresh ones. However, this frequently yields photos of poor quality. At several points during the image production process, GFP-GAN leverages a pre-trained version of an existing model (NVIDIA's StyleGAN-2) to inform the team's own model. Because of this, people's identities in images are retained.

However, this identifies a few flaws in the strategy. Sometimes the outputs are unusual and the resulting photos may not be particularly crisp, but identification can still slightly change. It is just impossible for us to guarantee that the rebuilt image will match the original. If we're lucky, the image will resemble our grandfather exactly, but Louis warns that it might as well resemble a stranger. Even though the outcomes are outstanding, users of these kinds of AI technologies should bear this in mind. You may download GFP-GAN from GitHub and give it a try.