The technology already exists that can reconstruct and alter images based on surrounding pixel colors and patterns, but the images that those programs created were often unrealistic and had many flaws, and required a decent amount of manual editing to make the pictures look real.

Nvidia, a U.S. based company, made a new program that is far more effective in accomplishing this, and nearly eliminates all need for manual touch-ups to the altered images.  The process includes many of the same features of previously developed programs, but at a much faster rate as well as a more realistic finished product.  The algorithms use machine learning to find the similarities and differences between facial expressions, scenery, animals, and so on.  Nvidia’s AI also picks up on patterns and colors far better than other similar programs.

Below is a comparison of the inputted image, another inferior program, Nvidia’s AI, and the original image.

As you can see, Nvidia’s AI is significantly better than the previous make, and the images it produces are very close to the originals.  Little to no manual editing is necessary.

While this type of technology does have the potential to to make a productive positive contribution to society, it has also been known to produce images that create political controversy, pornographic images, and photographs that can mislead research and scientific discoveries.  The way that this technology is used, and regulated, will be very interesting to watch unfold throughout the next few months.

https://www.sciencealert.com/nvidia-neural-network-editing-program-for-irregular-inpainting