A video showing actress Rashmika Mandanna entering an elevator made rounds on social media recently, sparking conversations about identity theft and artificial intelligence. The video was discovered to be a ‘deepfake’ where the actress’s face was transposed onto the body of a British-Indian influencer.
Rashmika took to X (formerly known as Twitter) to express her disappointment. She said, “Something like this is honestly, extremely scary not only for me, but also for each one of us who today is vulnerable to so much harm because of how technology is being misused.” Union Minister Rajeev Chandrasekhar reacted to the video and expressed his concern.
He stated that these dangerous and damaging forms of misinformation need to be dealt with by platforms. Rashmika’s experience raised debates about the misuse of AI and weaponising technology to harass women on the internet.
Simply put, deepfakes use a form of artificial intelligence called deep learning to make images of fake events. Not limited to videos, this technology can also be utilised to manipulate photographs and to deepfake audio, creating “voice skins” or “voice clones” of public figures. While some people use deepfake technology for political satire and comedy, the majority of deepfake videos on the internet are pornographic and usually involve celebrities.
With the number of deepfake videos on the internet growing exponentially and the technology constantly developing, it has become increasingly difficult to detect deepfake videos with the naked eye. However, despite the apparent believability of these videos, there are still some subtle imperfections and inconsistencies that can help to verify their authenticity.
Poor-quality deepfakes
Understandably, poor-quality deepfakes are easier to spot. You can look for flickering around the morphed faces as well as for patchy skin tone. Inconsistent facial features and disproportionate body parts such as the wrong length of arms, feet and forearms indicate that the image has been tampered with.
Unusual facial features
In deepfake videos, pay special attention to the moving parts of the face. Abnormal eye movements including little or too much blinking as well as gaze patterns that are not coordinated with the person’s actions are indicators of deepfakes. AI sometimes has difficulty in rendering very fine details like hair strands, teeth and reflections on the iris. Unusually smooth or wrinkled skin is another give away.
Lighting and colour
Inconsistencies and mismatches in the lighting, background, contrast and colour help identify deepfakes. Keep an eye out for irregularities in the lighting on the subject’s face and their surroundings, changes in brightness from one frame to another as well as subtle changes in skin colour and shadows.
Audio-visual mismatch
Deepfaked videos usually use AI-generated audio that has subtle imperfections. Comparing the audio and video quality can help to detect digitally manipulated content. Bad lip-syncing is also an important giveaway. Pay attention to whether the movements of the mouth correspond to the audio.