How Can You Spot a Deepfake?

Last week, I discussed the technology used to make deepfake media (DFM). This week, I want to discuss how to identify DFM. 

In the early stages of DFM creation, it was fairly easy to spot it. If the DFM was created through a face swap, several issues identified deepfakes. Major flaws were easy to spot (think: the fake face appearing like a mask or having distorted facial features). Minor issues were harder to spot, but they were (are) more common (e.g., deepfake videos had no blinking). Unfortunately, as people became more aware of DFM and how to spot them, the algorithms used to create DFM became better at creating realistic fake media. Now, obvious signs of DFM are mostly present in low quality or poorly made media, and more discreet signs of DFM are avoided by most high-quality algorithms.

Presently, DFM is most easily identified by unnatural movement, specifically in the eyes, lips, and overall body. Coloring issues are also somewhat easy to spot, especially if the lighting of the face does not match the lighting of the body. Natural-looking hair and teeth are hard to randomly generate via an algorithm, so it is common for DFM to lack individual strands of hair or individual teeth. Generally, if the image or video seems unnatural or “off,” there is a good chance that you should not trust that medium. 

Luckily, popular Internet sites where many users find news, like Facebook and Twitter, are working to ban the use of DFM in an attempt to prevent the spread of fake news. Next week, I will discuss the impact of DFM when it is not recognized or caught.