Instead of changing faces (like James Bond to Putin) what if, given sufficient training data, I wanted to:
- Remove or add some windows from a brick house?
- Convert a glass of red wine to a glass of white wine?
- Remove the infamous the Starbucks cup from Game of Thrones?
Although deepfakes are notoriously hungry to train, the pre-trained weights are plug-and-play.
Does there exist any equivalent for non-face objects?