A picture was posted on Reddit that immediately sparked controversy among old-school gamers. With the help of artificial intelligence, enthusiasts have redrawn the original rough pixelated image of 1993 - the face of Doom Guy, the protagonist of the game of the same name - into a photorealistic picture in high resolution. Many found the resulting face too kind and not at all as brutal as they had imagined three decades ago.
If the artists from the 90s were faced with the task of pushing the imagination of the audience, forming a unique image with the help of a handful of pixels, then modern AI has taken up the exact opposite task: drawing details. And it turned out to be much more difficult than the authors of the project thought. They had to run the initial meager image through several neural networks and still the result was far from perfect.
In the first phase, the pixel image was entrusted to be processed by AI applications such as FaceApp, Waifu2x, and GIMP. The picture was saturated with details, but it remained a pixel mosaic, so the StyleGAN neural network came into play. AI from Nvidia, after many iterations, managed to redraw the picture into a photorealistic image, but then it turned out that the original had the proportions that were initially incorrect for a human face. I had to edit Doom Guy's face manually in the editor.
However, the authors of the venture were not upset, and instead carried out the procedure of redrawing the pixel picture into a photorealistic one several times in order to draw up a clear algorithm. And now AI began to work on it quickly and efficiently, albeit with a large number of steps. For example, it was worth drawing Doom Guy with a series of separate pixels instead of compressed lips, and the neural network obediently provided the hero with a wide "Hollywood" smile. It was she who quarreled the gamers - a stern space marine should grin wryly, as in the original, and not this one.