I think you're describing pretty much what we are planning to do. Something we are looking into is actually just training the AI on a single mannequin head for all characters, since if it trains on a default set that never changes, we only have to weight the facial adjustments on one side of the transfer. That may not be possible, but if it is, it could save a huge amount of time spent adjusting facial alignments on each character.
This video is badly made, but it does show what our characters will look like post process. This is metahuman creator plus deepfake, and It's not a problem to train in makeup or demon face or whatever. For a final product, I'll want to go further than this guy did, and really make the tech work smoothly. It doesn't look bad here though, just right out of the box with minimal effort.
Maybe when I'm ready to test this out in a full scene, I'll make a video where Bruce Lee fights The Harlem Globetrotters on top of cars in moving traffic, like the matrix interstate fight scene with the agents jumping from car to car.
This video is badly made, but it does show what our characters will look like post process. This is metahuman creator plus deepfake, and It's not a problem to train in makeup or demon face or whatever. For a final product, I'll want to go further than this guy did, and really make the tech work smoothly. It doesn't look bad here though, just right out of the box with minimal effort.
Maybe when I'm ready to test this out in a full scene, I'll make a video where Bruce Lee fights The Harlem Globetrotters on top of cars in moving traffic, like the matrix interstate fight scene with the agents jumping from car to car.
Last edited: