Anyone ever watch those bad lip reading videos? Some are dumb, others are just hilarious. Even with years of doing ADR, I have no idea how these guys come up with the lines that seem to match the sync cues so perfectly.
The technology could be good enough now that they can manipulate the video details of her mouth moving. You’d only need a few dozen “virtual” emulations of facial muscle movements of vowels and consonants for the English language. They probably “sampled” those from stock videos, then programmed the sequences.
There were multiple versions of the “drunk Pelosi” video going around, which people kept changing to keep Facebook from deleting them all. We’re literally entering an era of such vast bizarre technology and “AI” that you won’t be able to tell what is real and fake anymore.
Honestly it’s not that difficult - and I doubt that technology plays much of a role. Just turn off the sound, watch the lips and start making up shit to match their movement.
Pretty soon you won’t even need video… just a still frame.