A friend who works in TV VFX pointed out to me that the deepfake only looks this good due to the base already looking pretty close to Hamill to begin with. So it's def not as simple as just slapping a young Hamill face on a stand in like a lot of people think. Even when you look at other deepfakes that were posted ITT they still look off, even more so compared to this one.
Of course, the existing VFX has given it a head start. Mainly with the shape/proportions. There's a lot of work with the overall head done besides just the face. It's more involved than just pasting a face on top of another actor. Practically the whole head gets replaced because their proportions aren't going to be the same. But if Hollywood incorporated technology like this
into their workflow it could have great results.
I don't mean to imply that they should completely abandon
all other vfx work to do
exclusively deepfaking and calling it done. I don't know why whenever someone complements deepfake technology, some people act like that means abandoning the
entirety of the vfx industry, as if they're just going to hit render and quit. It would be great if they used this in their
process. It wouldn't replace the
entire process, but it would replace and/or augment
important parts of it.
Deepfake doesn't do the
head, it only does the
face. They'd still have to do the work with the overall head and neck, plus blending and a lot of other tweaks, but instead of animating and rendering an emoting cgi face the traditional way, they could let the deepfake animate with the set actor's actual performance for more natural movement. Then put the rest of their bandwidth into blending that deepface layer into the cg head replacement that goes around/behind it, and fixing any artifacts or other unwanted quirks from that process (since the automation is rarely perfect).
For instance,
here the same channel did a deepfake of Ford in Solo. There was no underlying cgi to give it a head start like with Luke, so it's a more fair assessment of the tech's performance. The face and animation quality are still good, I believe that's a usable face layer to start with. However while it looks good, obviously the head/face proportions are wrong. That's where the more traditional vfx work would come in. And putting these tools
together could save them the trouble of manually recreating the face in every scene, while also producing a better final result.
Adding a new tool to the box doesn't mean all the other tools stop existing.