The Rise of Deepfake Videos: What You Need to Know
SUS IT
Understanding deepfake video technology, its implications, and how to protect yourself from synthetic video content.
Deepfake videos have gone from a niche curiosity to a mainstream concern in just a few years. With tools like Sora, Kling, and Runway making high-quality video generation accessible to anyone, understanding what deepfakes are and how to spot them has never been more important.
What Are Deepfakes?
Deepfake videos use artificial intelligence to create realistic-looking footage of events that never happened. This can range from face-swapping — placing one person's face onto another's body — to fully synthetic video generation where every pixel is AI-created. The term "deepfake" comes from "deep learning" and "fake," reflecting the neural network technology that powers these creations.
How Good Have They Gotten?
The quality of deepfake videos has improved dramatically. Early deepfakes from 2018-2019 were often easy to spot: faces would glitch during head turns, skin textures looked waxy, and lighting was inconsistent. Modern deepfake generators have largely solved these problems. AI-generated videos from tools like Sora can produce photorealistic scenes with accurate physics, natural lighting, and convincing human motion.
The Real-World Impact
Deepfakes have been used in financial fraud (fake CEO video calls authorizing wire transfers), political disinformation (fabricated speeches by public figures), revenge content, and social engineering attacks. In 2025 alone, reported deepfake-related fraud exceeded $25 billion globally. The technology has also created legitimate concerns in journalism, where video evidence — once considered reliable — can now be fabricated convincingly.
How to Spot Deepfakes
Despite improvements, deepfake videos still leave detectable artifacts. Here's what to look for:
Temporal inconsistencies: Watch for frames where the subject's appearance shifts slightly — a momentary change in skin tone, eye color, or facial structure that quickly corrects itself. This happens when the AI model produces inconsistent outputs between adjacent frames.
Unnatural blinking: Many deepfake models still struggle with natural blink patterns. Either the subject blinks too regularly (like a metronome) or doesn't blink enough. Real humans blink irregularly, typically 15-20 times per minute with varying duration.
Edge artifacts: Look carefully at the boundaries where the face meets the hair, ears, and neck. Deepfakes often produce subtle blending artifacts in these transition zones — slight blurring, color mismatches, or wavering edges that wouldn't exist in genuine footage.
Audio-visual sync: If the video includes speech, pay attention to lip sync accuracy. Deepfake generators sometimes produce lips that are slightly ahead of or behind the audio, especially on consonant sounds that require precise lip positions.
Forensic Detection
Beyond what the human eye can catch, forensic tools analyze deepfakes at the pixel level. SUS IT examines frame-by-frame consistency, noise patterns, compression artifacts, and temporal coherence across the entire video. These mathematical analyses can detect manipulation even when the video appears perfectly natural to human viewers.
Protecting Yourself
Be skeptical of sensational video content, especially if it appears to show public figures saying or doing something unusual. Check the source — was it posted by a verified account or a new, unverified one? Look for the video on multiple independent sources. And when in doubt, run it through a forensic analysis tool. In an era where seeing is no longer believing, verification is essential.