I wonder what will happen when somebody combines a GAN with a feature recognizing network like the Tesla cars use, so it can use its own extrapolated map of the surroundings to stabilize its output as the camera moves around.
I think it will have its limits, but the possibilities for editing together, supplementing and modifying products from smaller AI modules should stretch out what you can do on a small budget.
Imagine a future where people can compile written scripts into Hollywood quality movies.