Spoiler Alert: This post contains a minor spoiler for Rogue One, discussing something that is already fairly well known about a character in the film. It will not discuss plot details or anything like that. But if you're completely paranoid, stop reading now.
So it turns out that the rumours were true: Peter Cushing's Grand Moff Tarkin was one of the major characters in Rogue One, despite the fact that Cushing has been dead since 1994.
The CGI recreation in the film is very good. It steps a little bit into the uncanny valley, but is the best recreation like this we've ever seen. So how on earth did Disney's technical wizards manage it?
Some clues to how it has been achieved could perhaps be garnered from the public videos of Disney's research unit, which - perhaps not coincidentally - have posted a number of videos over the last couple of years demonstrating how faces could be recreated and manipulated in CGI.
The first challenge is going to be building a 3D model of Peter Cushing. Luckily, Disney has plenty of footage of him they can use. And in this video - "An Anatomically-Constrained Local Model for Monocular Face Capture" - you can see how Disney has the tech to take a 2D video recording and build a fully 3D model, by analysing the point data. Later in the video, it also shows how the virtual face can then be manipulated to create new expressions - they make the guy grin at one point. This is exactly what you'd need to do if you wanted to make Tarkin say new things.
Once the 3D model has been built, they will need a way to capture the new actor's... acting... and apply it to the 3D model. This video - "Synthetic prior design for real time facial capture" - shows how it can be done. Crucially, and perhaps importantly for actors, it appears that they can simply use a GoPro attached to a helmet - which means that they conceivably could have had the new guy act on the set - helping capture a better performance.
Making the 3D model as detailed as possible though is always is going to be tricky - but Disney has even built technology that will build a 3D model of a face from a camera, and then build a more detailed model that will predict where wrinkles will appear in the skin.
Hell, Disney has even spent ages working on how to make digital eyelids realistic.
And I'm not quite smart enough to work this out, but this video shows how mouth movements can be analysed for what sounds can conceivably be made with the lips in a given position - perhaps an indication on how the re-dubbing was achieved.
Watching these videos back in light of Rogue One, it certainly seems that these techniques could have been the ones used. And it also suggests that Disney's Research could be worth watching for subtle indications of what we can expect in future Star Wars films too.