For the last time, can we please just make this whole thing go away? Wait — you thought we were talking about the big meme-worthy saga surrounding Henry Cavill’s mustache? No way, why would anybody wanna do that? We’re talking about Henry Cavill’s actual mustache.
In a good-natured jab at the obsessive lengths to which major studios will go to unleash gigantic effects budgets on even the minutest of problems, one member of an online tech collective has demonstrated how to get rid of the world’s most infamous ‘stache — all for the price of some geeky elbow grease and a little well-placed faith in the magic of machine learning.
A member of the online “DeepFakes Club” community posted a YouTube video comparing the side-by-side results of Warner Bros.’ mustache-erasing efforts against those of a “neural network” of trained computers. In the process, the clip demonstrates that Justice League might have achieved the same result for Cavill's Superman — perhaps even a better one — simply by tapping the low-cost power of AI automation.
We could try to explain how it all works, but it’s probably better to stand aside and let the anonymous YouTuber — a person who’s clearly sitting around just waiting for the tech gods to raise the level cap on this kind of geek proficiency — take the wheel.
“Justice League had a budget of $300 million. I used a $500 computer bought off Craigslist and free deepfakes software to teach an AI how to digitally remove Henry Cavil’s mustache in a variety of situations. I also attempted to restore Superman’s correct face.”
Then the digital shaver goes on to say:
“…While deepfaking usually involves swapping two faces…I instead trained the computer model on two different versions, mustached and shaven, of the same face. The model did a decent job of removing Cavill’s mustache, at least as cleanly as the average rushed morning shaving session.”
If that explanation leaves you wanting a more technical description of how the borg-like face-swapping technology works, well … we’ll try. Essentially, deepfaking employs the speed and processing power of computers to do, en masse, what typically can only be done one image at a time with an editing program like Photoshop.
Think of it as a modern-day animation flip-book of still images, pegged to a whiplash frame rate that no human eye could ever detect. A skilled user sets the basic parameters, and then trusts the network to fill in the gaps as it learns on the fly. As you can see from the clip, the results are impressive.
“Traditionally, Photoshop and similar tools were used to make fake photos one at a time,” the collective’s website explains. “Deepfaking automates this process so that you can create convincing fake photos in large numbers. By combining large numbers of photos, you can generate a video. The automation is handled by deep learning using an artificial neural network.”
What's an "artificial neural network?" We're still a little, ahem ... fuzzy ... on that. But one thing's for sure: Thanks to these mustache-banishing geeks, Henry Cavill's face most assuredly isn't.
(via Heroic Hollywood)