Another chapter in our series "Who still doesn't believe that Artificial Intelligence (AI) will completely change the film industry?". This time it is a research project in which an AI model splits a scene into objects without any further user interaction. For example in trees, streets, cars, signs, people etc. These objects are then displayed colored in an object display. After that, you can now simply swap, enlarge, expand or replace the objects in an image. The following video shows how to imagine this in concrete terms:
You can easily replace one car with another, add trees or replace the road surface. And all this without any idea of compositing or image editing. The results are by no means perfect, but the whole thing is taking place on a photorealistic level.
A particularly large number of shortcomings can be seen when you take a closer look at the small details. Perhaps this could also provide a additional argumentation basis for 8K acquisition in the near future. Scaled down from 8K to FullHD, these synthetic details can hardly be identified as incorrect.