Analyse the Face and Eyes
The eyes and face are often the first places a deepfake will show its flaws. Watch for unnatural eye movement, such as a lack of blinking or a strange, fixed stare. Sometimes, the blink rate can seem robotic or out of sync. Look closely at the skin; it might
appear overly smooth and lack natural pores, or conversely, be blurry around the edges of the face and hair. Shadows and lighting can also be a giveaway. If the light on the subject’s face doesn’t match the lighting in the rest of the scene, you should be skeptical. These inconsistencies happen because AI models struggle to perfectly replicate the complex physics of light and natural human features.
Listen for Audio Clues
Your ears can be just as important as your eyes. A cloned voice often sounds flat, robotic, or lacks emotional range, even when the person is supposedly discussing an emotional topic. Listen for unnatural pacing or pauses in speech, as AI models often don't replicate the natural rhythm and breathing patterns of a real person. You might also notice poor lip-syncing, where the words you hear don't perfectly match the movements of the speaker's mouth. Another red flag is audio quality that is either too perfect—lacking any background noise—or contains strange, glitchy sounds that seem designed to mask audio imperfections.
Observe Body Movement and Posture
While facial manipulation is common, deepfakes often struggle with rendering the rest of the body realistically. Look for stiff or awkward posture that doesn't seem natural for the person speaking. Head movements might appear jerky or disconnected from the body, as if the head is an object layered on top of a separate video. Pay attention to how the subject’s body language matches their words. If someone is delivering an angry speech but their body is strangely still or their gestures are repetitive, it could be a sign of manipulation. These mismatches between facial expression and body movement are common artifacts.
Examine the Background and Edges
The focus of a deepfake is almost always the person, meaning the creator may pay less attention to the surrounding environment. Look for flickering or distortion at the edges where the subject meets the background. Hair is particularly difficult for AI to render, so blurry or unnaturally moving strands can be a significant clue. The background itself might be unusually blurry or show strange warping. Also, look for inconsistencies in objects in the environment. For example, if a shadow cast by a person doesn’t align with a shadow cast by a nearby object, it suggests the elements were compiled from different sources.
Stop and Verify the Source
Perhaps the most crucial check has nothing to do with visual or audio cues, but with critical thinking. Before you even consider sharing a shocking video, ask yourself where it came from. Was it posted by a reputable news organization or an anonymous account with few followers? If a public figure appears to say something outrageous, check their official social media channels or see if any trusted news outlets are reporting it. If a video is designed to make you feel angry or scared, pause and consider the motive behind it. This simple act of verification is one of the most powerful tools against the spread of misinformation.
