When Sega launched 1993’s Virtua Fighter, there was truly nothing of its caliber that came before. It was praised as the first fully 3D fighting game that stood at the forefront of technological innovation in comparison to the rest of the competition in the genre.
In the age of AI made freely available to auto-generate images, voices, text, video, and so forth, it’s quite fitting to revisit Virtua Fighter with a new coat of paint.
Using the incredible tools of Stable Diffusion, a Twitter user by the name of Colin Williamson managed to completely remake the game’s graphics–going from the blocky polygons of time past to realistic modern 3D textures.
Normally, Stable Diffusion uses a model base of text to image. However, Williamson recreated the look and feel of the game using Automatic1111. The upgrade looks less than that of a graphical facelift and more like real motion capture actors portraying the characters in a scene.
Even though it looks as though the changes were implemented through the simple click of a button, getting the look right with powerful AI required significant tinkering.
In an interview with Ars Technica, William said he had to “describe the character, and img2img does its best…though the hardest part was simply figuring out how to describe the characters’ clothes.”
On top of telling the AI what the characters should look like, Williamson also implemented what is known as negative prompting. This is where you tell the AI to avoid certain features, such as making sure the characters do not have more than 5 fingers.