The more inherent image understanding an AI can develop, the more magic is possible afterwards. A nice example is the current paper by Simon Niklaus, which takes the popular Ken Burns effect to a new level.
Various well-known technologies have been combined and partially optimized. For example, Z-depth estimation is combined with object separation to obtain defined edges and to prevent objects from being "torn apart" in depth. The AI also helps to conclusively supplement missing image content that is exposed behind objects in motion.
The following video shows the technical background as well as some impressive examples:
In addition, there are even more successful results to study here:
Since Long Mai and Jimei Yang from Adobe Research also worked on the paper, one can certainly expect a corresponding 3D Ken Burns effect in commercial Adobe products in the not too distant future.