What is a Deep Fake?

What they are and how to spot them online

Deep Fakes are fake images, audio, and video recordings that sound and look real. The word deepfake itself is a portmanteau of deep learning and fake, which refers to the way that machine learning and artificial intelligence (AI) are leveraged to create these fakes.

Deep fake technology has been employed for lighthearted entertainment purposes, but it has also put words in politicians mouths, enabled social engineering scams, and even cast celebrity faces in pornographic content without consent.

An illustration of an AI mapping and creating a deep fake face.
Tim Gamble / EyeEm / Getty

How Are Deep Fakes Created?

It might sound like science fiction or Hollywood special effects magic, but deep fake technology is surprisingly accessible. The most common method is the face swap deep fake, which replaces the face of a person in a video with the face of a different person.

This face-swap method relies on feeding video data of the original person and photographs of the replacement face to an AI algorithm. The AI analyzes the original face and the new face, finds similarities, and maps the new face onto the existing video frame by frame.

The entire process of creating a deep fake can be automated by web-based services, but the AI algorithms are also available for download, allowing anyone with enough expertise to create deep fakes on their own home computer.

Similar processes can be used to create fake images that look real, animate still photos using movement data from an actor, and even realistically impersonate a person's voice. It all relies on artificial intelligence to break down input data, like photos, videos, and audio recordings, and then synthesize new media that's indistinguishable from the real thing.

A Few Deep Fake Examples

A lot of deep fakes are malicious, like putting words in a politician's mouth, or pornographic in nature, but the technology has also been used for more lighthearted means.

Here are some deep fakes you can check out yourself:

  • George Lucas and Harrison Ford: This deep fake video is just one in a series of humorous Collider videos that film talented impressionists and replace their faces with celebrities. In this example, George Lucas and Harrison Ford react to Baby Yoda from The Mandalorian.
  • The Rock deep faked onto The Rock: In this deep fake from Corridor Crew, they use deep fake technology to fix the uncanny valley depiction of Dwayne Johnson as the Scorpion King by feeding an AI video data of Johnson from earlier in the same movie.
  • Henry Cavill's lip fixed in Superman: VFX artists were forced to hide Cavill's mustache in some Justice League scenes with cartoonish results. This side-by-side comparison shows how much better a deep fake can do.
  • Arnold Schwarzenegger's face on Bill Hader: In this clip, Bill Hader impersonates Arnold's voice, and Arnold's face is deep faked by Ctrl Alt Face onto Hader's body. The effect is removed when Hader uses his normal voice and applied when he impersonades Schwarzenegger.
  • Joe Rogan's voice deep faked by AI: In this clip, audio from Joe Rogan's podcast was fed into an AI algorithm to create a realistic voice model capable of saying virtually anything.

How Do You Spot Deep Fakes?

As the technology advances, it has become increasingly difficult to spot the difference between a deep fake and the real thing. Some are easier to spot than others, but the best ones look and sound extremely real. You'll have an easier time spotting ones created by amateurs who aren't able to leverage a ton of experience and processing power, but you have to be on the lookout for telltale signs of AI interference.

Here's how you can spot a deep fake:

  1. Look at the edges of the face: If the AI wasn't provided with enough input data, the new face may be clumsily pasted on the original head, creating obvious lines of demarcation or resulting in a face shape that doesn't look right.
  2. Examine the hair: If you can see individual hairs crossing over the face, it's probably a real video. AI has difficulty generating realistic hair, and individual hairs usually won't be visible in AI-generated portions of a video.
  3. Pay attention to mouths and teeth: AI typically has trouble generating individual teeth, which can be easy to spot if the video is high enough resolution. Also look for any errors when the mouth is opening and closing.
  4. If the person looks away, pause the video: Deep fakes do really well when the person is facing the camera, but they can struggle when switching to a profile shot or if the person turns completely away from the camera.
  5. Be wary of unknown information sources: While it's possible for a trusted source to be tricked and to run a deep fake as if it were real, you're much less likely to be tricked if you stick to trusted news sources.