deepfake difference (1)

Written by Shahzeb Rizwan Raja 12:40 pm Articles, Current Affairs, Pakistan, Published Content

Deepfake or Real? Here’s How to Tell the Difference

Deepfake, as the name suggests, uses deep learning technology to create convincing yet fake digital media. In this article, Shahzeb Rizwan Raja enumerates specific characteristics that distinguish between an actual and a deepfake video and image.
About the Author(s)
+ posts
Mr Shahzeb Rizwan Raja is currently a high school student.

Taking the internet by storm, a recent video leak of Pakistan Tehreek-e-Insaaf’s (PTI) senator, Azam Swati, left many at sixes and sevens. Many are baffled and frightened at this new necromantic Deepfake technology that has sprouted. In such heightened times of cyberattacks on our country, influential personalities are becoming more cognizant of their activities and are desperately grasping onto their dignity.

Unfortunately, in a country like Pakistan where even fake, tailor-made media content could end a person’s career, and impose undue stress on them by social media— which blindly blames the victims without any inspection—we have to tread cautiously and pay heed to every detail.

There are a lot of intricate ways of scrutinizing a video to ascertain whether it is a deepfake video or not, which requires some level of skill in computers. However, for the general public, I will list some tips you could look out for in such elaborately edited videos.

Above are a pair of photos from metaphysics tests showing a deepfaked Jim Carrey, and the result of him turning to the side. You can catch the editing from the photo of the deformed side profile, and this is something you should be keeping an eye out for. However, this is a more visible form of deformation. In other videos, it could be extremely subtle and would need a more wary eye to locate it.

Also Read:  The Case for STEM Education in Pakistan

The reason for this deformation is that most deep-fake applications use 2D-based facial alignments algorithms, which assign only 50-60 percent of the number of landmarks from a front-on face view to a profile view. Without the algorithm being able to see enough reference points for the side profile, it simply does not know how to project that part of the fake face.

Besides this, there are 8 points that are highlighted in an article published on the Massachusetts Institute of Technology (MIT) media page regarding how to discern between a real and a deepfake video.

  1. You must pay close attention to the face because high-end deepfake manipulations are almost always facial transformations. 
  2. Pay attention to the cheeks and forehead, and notice if the skin’s texture and condition match the agedness of the hair and eyes. Deepfakes are often incongruent on some dimensions.
  3. Pay attention to the eyes and eyebrows, and check if there are shadows and lighting details. Deepfakes often fail to fully represent the natural physics of a scene. 
  4. Pay attention to the glasses and notice any changes in the glare. Once again, deepfakes often fail to fully represent the natural physics of lighting.
  5. Pay attention to the facial hair or lack thereof. Deepfakes might add or remove a mustache, sideburns, or a beard, but they often fail to make facial hair transformations fully natural.
  6. Pay attention to facial moles. 
  7. Pay attention to blinking.
  8. Pay attention to the size and color of the lips.

As deepfake video technology continues to advance, gauging differences between videos is becoming more serpentine. This is always going to be a game of cat and mouse. As soon as a new model to detect one type of deepfake will be in circulation, there will be a generator that will be able to beat that detector.

Also Read:  The War In Yemen: The Largest Humanitarian Crisis in the World

We, as a country, have to be up to date in making our people aware through widely used social media platforms, like Tiktok, Facebook, or Instagram, on new forms of digital attacks, and how to steer clear from them.

If you want to submit your articles and/or research papers, please check the Submissions page.

The views and opinions expressed in this article/paper are the author’s own and do not necessarily reflect the editorial position of Paradigm Shift.

(Visited 597 times, 1 visits today)
Click to access the login or register cheese