1. Examine The Face and Emotions
The first place to look for signs of manipulation is the human face. AI models, while advanced, often struggle with the subtle nuances of real human features and expressions. Pay close attention to the eyes; deepfakes may feature subjects who blink unnaturally,
too little, or not at all. The direction of their gaze might also seem off or disconnected from the action in the video. Look for skin that appears too smooth or too wrinkly for the person's age, almost like a digital mask. Other giveaways can include unnatural facial symmetry, or mismatches in skin tone between the face and the rest of the body. If the person’s emotional expressions seem flat, robotic, or don't align with the tone of what's being said or done, it's a significant red flag.
2. Look for Visual Inconsistencies
Beyond the face, the overall visual composition of a video can reveal clues. Deepfake generation often leaves behind subtle artifacts, especially around the edges of the manipulated subject. Look for blurring, flickering, or jagged edges where the person's hair or face meets the background. Lighting is another area where AI can falter. Check if the shadows and highlights on the subject match the lighting of the surrounding environment. If the lighting on the face seems inconsistent with the rest of the scene, or if shadows appear in unexpected places, it could be a sign of a composite image. Sometimes, asking the person in a live video to turn their head to a sharp profile angle can expose a deepfake, as the technology often struggles to render faces from the side.
3. Analyse the Audio for Flaws
Our ears can be just as important as our eyes in detecting fakes. Manipulated videos often feature cloned voices or poor lip-syncing. Listen carefully to the audio track. Does the voice sound robotic, flat, or lack emotional variation? AI-generated audio can have an unnatural cadence and may be missing the small, in-between sounds of a real human talking, like breaths. Also, watch the person's mouth closely. In many deepfakes, the lip movements are not perfectly synchronised with the spoken words. Any lag or mismatch between the mouth movements and the audio is a strong indicator of manipulation. Inconsistent background noise that cuts in and out abruptly can also be a giveaway.
4. Slow It Down and Go Frame-by-Frame
Many deepfake imperfections are too brief to be noticed at normal playback speed. If a video feels suspicious, one of the best techniques is to slow it down. Use a video player or an online tool that allows you to watch at a slower speed or advance the footage frame by frame. Doing so can reveal glitches, flickering, or morphing that would otherwise be missed. This is particularly effective for spotting inconsistencies around the edges of the face or in fast-moving sequences where the AI might struggle to keep up. Keep an eye out for strange artifacts or visual noise that appears for just a fraction of a second. This level of detailed inspection is difficult for even sophisticated deepfakes to withstand.
5. Always Question the Source
Perhaps the most crucial step has nothing to do with technology and everything to do with critical thinking. Before you analyse a single pixel, ask: where did this video come from? A video being shared widely does not make it authentic. Look for the original source of the clip. Was it posted by a credible news organisation, an official agency, or a verified account? Be wary of videos that appear without context on messaging apps or anonymous social media accounts. You can perform a reverse video or image search to see where else the content has appeared online and whether it has been debunked by fact-checkers. Often, the context and origin of a video are more revealing than any visual flaw.
