Analyse the Face and Eyes
The eyes are often the first giveaway in a deepfake. AI models still struggle to replicate natural, irregular blinking. A person in a deepfake video might blink too often, too little, or in a weirdly rhythmic pattern. Sometimes, they might not blink at all,
creating an unsettling, doll-like stare. Look closely at the rest of the face for inconsistencies. Is the skin unnaturally smooth or waxy, or do facial expressions seem disconnected from the person's tone? Other red flags include mismatched facial symmetry, where ears or jawlines don't quite align, and poor rendering of teeth. These subtle flaws often break the illusion of reality.
Inspect Hair, Hands, and Edges
Artificial intelligence finds it difficult to render fine details perfectly. Look for blurriness or strange artefacts around the edges of a person's head and hair. Individual strands of hair might appear to melt into the background, or the entire hairline can flicker from frame to frame. Hands are another infamous weak point for AI generators. You might spot inconsistencies like too many or too few fingers, unnatural bends in the limbs, or stiff-looking wrists. How a person interacts with objects can also be a tell. If someone is holding a phone or a cup, the way their hand grips it might seem physically awkward or incorrect.
Listen for Audio Mismatches
Deepfakes are not just a visual threat; audio can be cloned with alarming accuracy. However, there are still clues to listen for. A classic sign is poor lip-syncing, where the audio doesn't perfectly match the movement of the speaker's mouth. The voice itself might sound robotic, flat, or lack emotional range, even when discussing an emotional topic. AI-generated speech often has an unnaturally steady cadence and lacks the small filler words, pauses, or breaths a real person makes. Pay attention to the background noise as well. The audio might be suspiciously clean with no ambient sound, or you might hear sudden cuts in the background noise.
Slow Down and Check for Glitches
Deepfakes are designed to be convincing at normal speed, but flaws often appear when you look closer. Pause the video and go through it frame by frame. This can reveal strange visual distortions, especially where the manipulated face meets the neck or background. You might see a 'flickering' or 'melting' effect at the edges. Look for inconsistencies in lighting and shadows. Does the light on the person's face match the lighting in the rest of the scene? Reflections in their eyes or on glasses should also match their environment. Frame-by-frame analysis can also expose generative flaws like a piece of jewellery that disappears and reappears.
Verify the Source and Context
This is the most critical check of all. Before you even analyse the pixels, consider the source. Is the video being shared by a reputable news organisation or an anonymous account with a strange history? The rise of deepfake scams in India, from fake celebrity investment ads to impersonations of family members in distress, highlights the need for caution. Recently, police in Tamil Nadu even registered a case against AI-generated videos of the Chief Minister being used for a financial scam. If a video makes a sensational claim or tries to create panic, take a moment to search for corroborating reports from trusted sources. Often, a quick search is the fastest way to debunk a fake before it fools you or others in your network.
