7 min read
Deepfakes – the art of inserting someone’s likeness into a video, thereby creating an artificial rendition of a person – has been around for a while. Although some of the initial applications were dubious, being used to insert celebrities’ likenesses into pornography or to seemingly place someone into a compromising position, we are now moving into mass use of the technology for more positive use cases.
Recently, we have seen it applied in movies so actors can play younger versions of themselves, like Robert De Niro in The Irishman or even to alter facial features when Henry Cavill had a moustache removed during reshoots for Justice league: Dawn of Justice. In Rogue One: A Star Wars Story, a digitally created version of deceased actor Peter Cushing reprised his character from the original 1977 movies as well as a faked version of Carrie Fisher’s Princess Leia.
“Deepfake is actually a new name for researchers. Computer graphic and machine learning people have been trying to use computers to synthesize humans in a virtual world for some time,” says Min Sun, Appier’s Chief AI Scientist.
The Development of Deepfake
Since the technology started to develop about five years ago, Sun says we have progressed from 2015 when it was possible to imitate a face (but not the teeth) to real-time re-enactments by 2017 and now on-screen robots ‘driven’ by human actors.
Initially, these methods required thousands of images in order to create computer models which is why celebrities were a target for deepfakes. However, the technology has moved forward and now a deepfake can be generated from a single image.
“These real-time deepfakes require a ‘driver’ – someone who is playing a part. The ‘actor’s’ likeness is then superimposed over the driver to create a deepfake,” explains Sun.
For brands and marketers, the use of deepfake technology presents a great opportunity. With the COVID-19 pandemic, access to talent for spokesperson-led campaigns has been limited, but deepfake technology lets you use a local actor to play a part and then alter the footage and add the spokesperson in later.
It also allows creatives to have actors interact with younger versions of themselves. For example, a commercial recently used a deepfake depiction of the SportsCenter anchor Kenny Mayne. He is currently 60 but a younger version of him was used to make it seem he was prophetically talking about the present day.
The technology has evolved over the last two or so years, allowing the changes in videos to look extremely realistic and only discernible as a digitally altered fake to someone with a trained eye.
Credibility and Realism: Deepfakes in Marketing
Today’s deepfake technology can be very convincing, but there are critics saying it represents a threat to actors and a challenge to the estates of deceased actors.
For marketers, there is a different, but related problem.
Can we trust that the person on the screen is actually who we think it is? Is it important for viewers to be made aware when a “fake” actor is presenting a product to them? In some cases, it is obvious when the technology is used to allow a spokesperson to seemingly speak in multiple languages.
For a company looking to use a recognized personality speaking to market in their mother tongue, it can be a powerful tool. This was done to great effect with a campaign called Malaria No More that had well-known soccer star David Beckham speak nine languages.
Dior has “revived” Marilyn Monroe for an ad campaign, following on from Dove doing the same with Audrey Hepburn a few years before.
It is likely people employed their own suspension of disbelief. They knew Beckham was not fluent in their native language and that Monroe and Hepburn had passed away, but they accepted it.
The technology even lets historians play an exciting game of “What if…”.
Alan Kelly, from Dublin’s Rothco, learned that on the day John F Kennedy was assassinated, he was on his way to give a speech. By tracking down a copy of the speech and using stored footage and recordings of the President, Kelly was able to recreate the speech, giving voice to the man who had been dead for over five decades.
Although the early days of deepfake technology were plagued with bad publicity, the future is brighter. The ability to create consistent messaging across languages and geographies is something marketers and advertisers have been chasing for years.
Anthony Baker from R/GA Tokyo recently said, “deepfake technology is empowering the creation of localized and translated content, and light personalization without the high cost of filming multiple variations.”
As well as delivering consistency in messaging, it has the potential to reduce costs significantly.
Putting the Customer at the Center
For some time now, fashion retailers have been looking for ways to ‘dress’ customers virtually, allowing them to see how particular garments will look on them before committing to a purchase. That also allows sellers to make suggestions and create a more personal experience for the customer.
Deepfakes can take this further. Rather than using a static image, the customer can be put in the ad to have a more immersive experience with a product. Suddenly, they can see themselves alongside a celebrity, wearing the latest fashion, thus creating a more emotional connection to the brand and the products.
When it comes to supporting customers, we have seen the evolution of chatbots and voice assistants over recent years. But we are close to the day when you could make a video-conference call for help and have the celebrity that endorses the product respond and help over video.
Sun says a source person can be used to drive an actor’s likeness.
“A new thing you can do is take the source appearance and overlay on the target’s expression, as well as the outfit and the whole background. So, you actually swap the face. Instead of training a bot to respond in an actor’s likeness you have a human driver.”
This would allow the next generation of chatbots and voicebots to look like a trusted person.
Like any new or emerging technology, there is capacity for deepfakes to be misused. Just as selectively editing a politician’s words can alter the meaning of what they are saying, a deepfake can be used to corrupt a message or literally put words in someone’s virtual mouth. The potential for deception is significant.
Should the use of a deepfake be disclosed? Sun says people need to be aware that a video may not be true evidence. That means we need to be aware of the source of videos and careful about sharing videos as they can be misused. Fortunately, tools are being developed that use AI to detect deepfake videos.
For marketers, being clear when a deepfake video is being used is important so customers don’t feel they have been deceived.
Deepfake technology offers significant opportunities but its use should be tempered by ensuring viewers are not misled. That means any actor used in a deepfake needs to be aware of the use and what messages they are being used to deliver.
As the technology becomes democratized, with the likes of Snapchat and TikTok already looking at making it universally available, the potential for misuse is growing. When it comes to putting this technology into the hands of mainstream users, Sun says “I think right now, because some people are already building to a cell phone solution, it is currently doable as long as you have one person driving.”
With the increasing accessibility of deepfakes, there will be a temptation to use more and more, but like any new technology, it needs to be employed judiciously. Businesses that use deepfakes intelligently and honestly will be able to reap considerable benefits in personalization and customization for their customers everywhere from advertising and the sales process through to support.
* Want to learn more about how you can use AI to enable data-driven decision making and improve marketing efficiency? We are here to help! Get in touch with us today for an exclusive consultation.