- Joined
- May 6, 2019
- Messages
- 11,214
- Points
- 83
Not that the original Doomguy lacks charm (pixelated face and all), but someone has figured out how to leverage the power of a neural network and some sophisticated upscaling tools to draw the classic character's portrait in high resolution.
This is basically what Doomguy would look like in real life. Sadly, he's all covered up in the newer games.
The final image is actually the product of several AI systems taking the improving on one another. According to the Reddit thread, the original “Doom Guy” sprite was fed into the AI of photo-editing programs including FaceApp, Waifu2x, and GIMP. That resulted in an improved, but still very pixelated and maybe pinkeye-infected version of the face.
From there, the image was fed into StyleGAN, the Nvidia AI system that people have used to create photorealistic portraits and nightmarish Pokémon sprites. StyleGAN churned out a more realistic image then took another pass to smooth it out. But because the original sprite had unrealistic proportions, the new image was manually reshaped, into this final version.
This is basically what Doomguy would look like in real life. Sadly, he's all covered up in the newer games.
The final image is actually the product of several AI systems taking the improving on one another. According to the Reddit thread, the original “Doom Guy” sprite was fed into the AI of photo-editing programs including FaceApp, Waifu2x, and GIMP. That resulted in an improved, but still very pixelated and maybe pinkeye-infected version of the face.
From there, the image was fed into StyleGAN, the Nvidia AI system that people have used to create photorealistic portraits and nightmarish Pokémon sprites. StyleGAN churned out a more realistic image then took another pass to smooth it out. But because the original sprite had unrealistic proportions, the new image was manually reshaped, into this final version.