If you’re one of the billions of people who’ve posted pictures of themselves on social media over the past decade, it might be time to rethink that behavior. New AI image generation technology allows anyone to save a handful of photos (or video frames) of you and then train AI to create realistic fake photos that show you doing embarrassing or illegal things. Not everyone may be at risk, but everyone should be aware of this.
Photos have always been subject to counterfeiting – first in dark rooms with scissors and paste and then by pixels through Adobe Photoshop. But it took a lot of skill to come across as convincing. Today, making compelling photo-realistic fakes has become almost trivial.
Once an AI model learns how to represent someone, their image becomes a software toy. The AI can create images of it in infinite amounts. And the AI model can be shared, allowing other people to take images of that person as well.
John: A case study on social media
When we started writing this article, we asked a brave volunteer if we could use their social media images to try and train an AI model to create fakes. They agreed, but the results were too convincing and the reputational risk turned out to be too great. Instead, we used AI to create a set of seven simulated social media photos of a fictional person we’ll call “John.” That way we can safely show you the results. Now let’s pretend that John is a real dude. The result is exactly the same as you will see below.
In our fantasy scenario, “John” is an elementary school teacher. Like many of us, John has spent the last 12 years posting pictures of himself on Facebook at work, relaxing at home, or while going somewhere.
With nothing but those seven images, someone could train AI to generate images that make it look like John is living a secret life. For example, he would like to take nude selfies in his classroom. At night, John might go to bars dressed as a clown. On weekends he could be part of an extremist paramilitary group. And maybe he was in jail on an illegal drug charge, but hid it from his employer.
We used an AI image generator called Stable Diffusion (version 1.5) and a technique called Dreambooth to teach AI how to create images of John in any style. While our John isn’t real, someone could reproduce similar results with five or more images of a person. They can be taken from a social media account or even taken as still images from a video.