You have probably seen a Wombo video circulating on your social networks. Maybe it was Ryu’s Street Fighter singing the “Sorcerer” or the last three heads of the US Federal Reserve imitating in unison “Never Going to Give You Up”, by Rick Astley. Each clip features exaggerated facial expressions and mysterious, sometimes nightmarish animations. They are stupid, fun and offer a useful insight into the current state of deepfakes.
It is certainly getting faster and easier to make fakes generated by AI, but the more convincing they are, the more work is needed. Tom Cruise’s realistic deepfakes that went viral on TikTok, for example, required an experienced VFX artist, a first-rate impersonator and weeks of preparation to work. One-click fakes that can be created without effort and experience, by comparison, still look like those made by the Wombo app and will continue to be so in the immediate future. In the short term, at least, deepfakes will obviously be manufactured and an instant bait for memes.
The Wombo app was launched at the end of last month in Canada after a short development process. “In August 2020, I got the idea for Wombo while smoking a joint with my roommate on the roof,” said Wombo app creator and CEO Ben-Zion Benkhin The Verge. The product launch was “a huge joy,” he says. “I’ve been following the AI space, the meme space, the deepfake space, and I just saw the opportunity to do something cool.” In just a few weeks, Benkhin estimates that the app has seen about 2 million downloads.
Wombo is free and easy to use. Just take a picture of your face or upload an image from the camera roll and press a button to synchronize the image with one of the songs adjacent to the meme. The application software will work its magic on anything that looks vaguely like a face and on many things that don’t resemble each other. Although similar apps in the past have been haunted by privacy fears, Benkhin insists that user data is safe. “We take privacy very seriously,” he says. “All data is deleted and we don’t share it or send it to anyone else.”
The name of the app comes from sports slang, specifically Super Smash Bros. Melee. “If a player falls like a crazy combination, the casters will start shouting ‘Wombo Combo! Wombo Combo! ‘”Says Benkhin. True to these origins, Wombo proved to be particularly popular with players who used it to animate characters from titles like League of Legends, Fallout: New Vegas, and Dragon Age. “I did some research [the origins of the slang], ”Says Benkhin,“ and apparently there was a pizzeria that started it all, where they put a ton of topping on the entire pizza and called it the Wombo Combo ”.
Benkhin says the app works by transforming faces using predefined choreographies. He and his team filmed the base video for each song in his studio (“which is actually just a room in my apartment”) and then used it to animate each image. “We stole the movements from their faces and applied them to their photo,” he says. The app is also an example of the fast-paced world of AI research, where new techniques can become consumer products in a matter of weeks. Benkhin notes that the software is built “on the basis of existing work”, but with subsequent adjustments and improvements that make it “our own proprietary model”.
Currently, Wombo offers only 14 short music clips to sync, but Benkhin says he plans to expand those options soon. When asked if the app has the appropriate licenses for the music he uses, he refuses to respond, but says the team is working on it.
However, as with TikTok, it appears that the range offered by Wombo could help to alleviate license holders’ concerns about rights. Wombo has already been approached by artists who want to put their music on the app, says Benkhin, and this is likely to offer a revenue stream beyond the current premium level (which pays for priority processing and no in-app ads). “Will, give [artists] a completely new way to engage the public, ”he says. “It gives them this new viral marketing tool.”
Wombo is far from the first application to use machine learning to create fast and fun deepfakes. Others include ReFace and FaceApp. But it is the latest example of what will become an increasingly prominent trend, as deepfake apps become the latest meme models, allowing users to mix favorite characters, popular songs, choreographed dances, public figures and more. The future of deepfakes will definitely be memeified.