It wouldn't be compulsory to give it an actual AI, could just be a chat bot with text-to-speech and speech-to-text (probably free open source libs availabe).
I mean just look at Eliza, (ripped from wiki)>Weizenbaum regarded the program as a method to show the superficiality of communication between man and machine>However, many early users were convinced of ELIZA’s intelligence and understanding, despite Weizenbaum’s insistence to the contrary.
You take it in small steps, learn as you go. Don't need the "waifu" to be free-walking to begin with, do like those simple card games and such that spawn a 3D model (seen through phone AR app) from a card thingy, conclude orientation of the surface you put it on by the angles of the card edges. First iteration would just be to spawn your image T-posing right at the camera without correcting anything, size or angles, just a waifu "hello world" to overlay something over video.>Advanced hardware
Why would you use advanced hardware to prototype amateur stuff?
Just prototype it on a smartphone.
If you think you won't have to make effort, yeah of course it won't happen. You won't get results alluded to in the thread. But it's a start, and if you keep at it maybe you'll be the first to actually have something in that area when you apply what you learnt.
Everything is made by people.
That doesn't mean they didn't make an effort to get there.