The other two factors, resolution and quality, are related but different. Resolution refers to the number or size of pixels in an image, while compression is the process of reducing the size of a video file by discarding detail, often leaving behind blocky patterns and blurry edges.
In fact, Farid says that low-quality fakes are so convincing that bad actors will intentionally lower the quality of their work. “If I’m trying to fool people, what do I do? I generate a fake video, lower the resolution so it’s still viewable but every detail is understandable, and then add compression to further obfuscate any possible artifacts,” Farid says. “It’s a common technique.”
Serenity Strall/Getty ImagesThe problem is, even as you read this, tech giants are spending billions of dollars to make AI even more realistic. “I have to give you some bad news: If these visuals are here now, it won’t be anytime soon,” Stam says. “We expect these visual cues, at least the obvious ones, to disappear from videos within two years, because they’ve already largely evaporated from AI-generated images. You can’t believe your eyes.”
That doesn’t mean truth is a lost cause. Researchers like Farid and Stam have more sophisticated technology at their disposal when verifying content. “When we generate or modify a video, it leaves a tiny statistical imprint that is invisible to our eyes, like a fingerprint at a crime scene,” Stam says. “We’re seeing the emergence of technology that helps us look for and reveal these fingerprints.” For example, the distribution of pixels in a fake video may differ from the real video, but such factors are not certain.
