Previously, we featured an AI that could create photorealistic images of apples from oranges, or zebras from horses, and vice versa. Now, we don't even need a photorealistic base image, just a basic segmentation map outlining shapes to create a video from it. Previously this wasn't possible because AIs would have trouble simulating the progression of movement over time, but with the help of a new AI algorithm that could understand the rules of Pac Man after watching only 120 hours of play, this changes everything!
"World-Consistent Video-to-Video Synthesis" by Arun Mallya*, Ting-Chun Wang*, Karan Sapra, and Ming-Yu Liu
https://nvlabs.github.io/wc-vid2vid/
