With a general election looming, a ‘deepfake’ video of Boris Johnson endorsing Jeremy Corbyn and vice versa has been grabbing a lot of media coverage this week. Suddenly everyone seems to be talking about deepfakes but does the tech open up a new world of creative possibilities for brands or pose the biggest threat they’ve ever faced?
For the uninitiated, a deepfake is an AI-generated video in which the protagonist is either completely replaced or manipulated to say or do whatever the creator wants. The technology is hugely controversial due to the ethical questions it raises and its potential to cause harm. Just imagine if creating a fake video suddenly becomes as easy as creating a piece of fake news. Deepfakes can weaponise misinformation and could be used to sway an electorate, discredit a business rival or even break up a happy marriage. It’s all scary stuff but it can also offer huge creative opportunities for brands when used responsibly.
Hollywood has used techniques such as face swapping and de-aging for many years now and recent films like Gemini Man and The Irishman have shown how far the technology has evolved. These effects have traditionally been achieved on big budgets through a complex array of head-mounted cameras and lengthy post-production work. However, publicly available software featuring cutting edge AI technology is now empowering creatives to achieve similar results from the comfort of their homes.
YouTuber Ctrl Shift Face is a renowned ‘deepfaker’ who specialises in face-swapping actors in iconic movie scenes. As you’ll see below, his work rivals some of the best effects that Hollywood can offer.
So how does it work? The process requires two videos: i) the original video that you wish to doctor; and ii) a video of the protagonist that you wish to insert into the fake. These videos are then fed into a neural network - a set of algorithms structured like a human brain which is ‘trained’ to recognise different patterns. The neural network extracts each video frame and analyses them for details such as the position of facial features, the movement of the mouth, angle of eye gaze, etc. This process takes a few days to complete, after which the resulting data is used to merge the two videos together to produce the final deepfake.
The technology is evolving all the time, with impressive results now being achieved through mobile apps. The Chinese app Zao recently created a lot of buzz as it enabled users to comp themselves into famous movie scenes by simply uploading a single facial image. The resulting video was rendered in less than 8 seconds and was of a very high quality. Sadly, the app was plagued with controversy due to data privacy concerns over the ownership of the uploaded images and generated videos.
Despite the obvious moral, legal and ethical questions, brands are starting to use the technology to market their wares. The charity Malaria No More used a technique known as audio-visual synthesis to create a video in which David Beckham appears to speak nine languages using the voices of malaria survivors and pioneering doctors from around the world.
Deepfake technology was used to resurrect artist Salvador Dali for the Dali Museum in Florida.
Savvy brands have long embraced the power of social media to connect audiences through celebrity endorsements and other forms of influencer marketing. However, such activity is relatively costly and tend to be limited to a very specific marketing message. The power of deepfake technology can enable brands to reach many different customers with highly targeted, potentially personalised messaging at a fraction of the cost. Just imagine your favourite actor, sport star or musician delivering a marketing message just to you, where they refer to you by name and the content is perfectly tailored according to your profile data. This is the next level in influencer marketing and I’m sure it’s just round the corner.
The technology also brings potential benefits to production as late script changes will no longer require costly re-shoots as talent could be pre-scanned to enable quick edits or even further creative development without their direct involvement. As per the earlier Dali example, the technology can even be used to resurrect the dead to ensure that their legacy lives on. You can easily imagine a scenario where brands no longer have to pay for talent time, but instead secure usage rights for their digital likeness.
Perhaps the most exciting application of the technology is the ability to seamlessly insert the user into the ad creative they see online. They will effectively become the stars of these ads, possibly leading to greater brand affinity and increased loyalty. Such executions will be subject to users granting access to their data but this is a natural progression on what the likes of Google and Facebook are already using today.
As with all technology open to misuse, brands must show an ethical responsibility - not just to the subjects of deepfake videos, but also to audiences - to provide education and transparency around the videos. In this era of fake news, it’s particularly important for brands to indicate that their video harnesses deepfake technology. This can be achieved through a simple disclaimer that displays at the start of the video or through a stamp of trust like the ‘blue tick’ used for official accounts on Twitter, Facebook and Instagram. Whatever method is used, it’s important that brands are fully transparent otherwise they risk damaging their reputation and losing customers.
So what does the future hold? Perhaps we’re on the cusp of a new era where the tech will empower brands to really push the creative envelope to deliver campaigns that could only have been dreamt about before. Others will argue that the tech will sound the death knell for brands, rendering celebrity endorsement and influencer marketing as meaningless. Consumers will no longer know who to trust or believe that a marketing claim is genuine.
Realising the threat to their precious ad revenue, the likes of Google and Facebook have already spent millions developing sophisticated deepfake detection algorithms. However, despite claims of very impressive detection rates, deepfakes will always find an audience due to the dynamics of the modern web where sharing is frictionless and attention is monetised.
Either way, deepfakes aren’t going away and will only become more sophisticated. Brands will need to tread carefully but if they’re able to strike the right balance between consumer trust and creative innovation then the rewards will be there for them.
Bharat Trivedi, Technical Director