Real Person or AI-Generated Video?
AI can now generate hyperrealistic videos of people — walking, talking, moving naturally. Two clips per round. Can you tell which person is real?
Free · No account required · Train your deepfake detection
How the Deepfake Video Quiz Works
Each round shows you two short video clips side by side — one shows a real person captured on camera, the other is AI-generated. Your job: pick the real one.
After you answer, both videos keep looping so you can study the telltale signs — subtle facial glitches, unnatural skin texture, off-timing blinks, or hair that behaves slightly wrong.
This is one of the hardest categories in the quiz. Current AI video models for human subjects are remarkably convincing. Most first-time players are surprised by how difficult detection has become in 2026.
Which AI Video Models Are Used?
The quiz features the most advanced text-to-video models tested specifically on human subjects:
Kuaishou's flagship model generates realistic human movement and expressions. Kling 3.0 is particularly convincing on walking, gesture, and upper body motion — making it one of the most deceptive person generators available today.
OpenAI's Sora 2 handles complex human scenes with realistic physics and spatial coherence. Strong at consistent identity across frames and natural interaction with environments.
Google's Veo 3 produces cinematic human footage with exceptional skin rendering and lighting. Its strength is in slow, controlled movement — walking, sitting, gesturing — where it approaches photorealism.
ByteDance's model produces smooth, natural-looking human motion with high temporal consistency across frames. Strong at full-body movement and casual everyday actions.
See all models compared: Full AI video model comparison →
Why Deepfake Detection Matters in 2026
Deepfake technology has advanced to the point where AI-generated videos of people are nearly indistinguishable from real footage for most viewers. In 2025 and 2026, deepfakes have been used in financial fraud, political misinformation, identity theft, and non-consensual content at scale.
Training your eye to recognize the subtle artifacts that current models still produce is one of the most practical media literacy skills you can develop. It is also increasingly relevant for journalists, HR professionals, law enforcement, and anyone who evaluates video content as part of their work.
This quiz does not use detection algorithms or AI classifiers — it trains your human visual perception directly, which is how most real-world deepfake encounters happen.
How to Spot AI-Generated Videos of People
Even the most convincing deepfakes leave traces. Here is what experts look for:
Hands remain the most reliable deepfake tell. Watch for fingers that merge, bend at wrong angles, briefly multiply, or look anatomically inconsistent during motion. This is especially visible during grasping or gesturing.
Individual hair strands tend to merge into solid masses, float slightly off the head, or behave with too little physics response. Real hair responds differently to every movement and frame.
AI blink patterns are often too regular or too infrequent. Eye tracking can feel mechanical, with the gaze drifting in slightly unnatural arcs. Real eye movement is irregular and reactive.
AI skin can look overly smooth, plasticky, or exhibit subtle flickering — especially at the edges of the face and around the hairline. Real skin has pores, micro-imperfections, and consistent texture under all lighting.
Fabric wrinkles and folds may not move realistically with body movement. AI clothing sometimes stretches, flickers, or fails to react physically to motion in the way real fabric would.
AI backgrounds sometimes drift, shift, or briefly distort — especially near the edges of frame where the model pays less attention. Real footage has stable, physically consistent backgrounds throughout.
Frequently Asked Questions
How realistic are AI-generated videos of people in 2026? ▾
Very realistic. Models like Kling 3.0, Sora 2 and Veo 3 generate person videos that fool most viewers on first viewing. Hands, hair, and eye movement remain the most reliable tells, but even these are improving rapidly with each model generation.
What is a deepfake? ▾
A deepfake is an AI-generated video of a real or fictional person. Modern deepfakes use text-to-video models that generate entirely synthetic footage — not just face swaps. The result can be a completely artificial person performing any action described in a prompt.
Is the quiz free? ▾
Completely free with no account required. Play unlimited rounds directly in your browser on any device.
What is the average score on this quiz? ▾
The person video category is the hardest in the quiz. Most first-time players score around 55-60% — barely above random chance. After learning what to look for, scores typically improve to 70-75%. Consistent accuracy above 85% indicates a well-trained eye.
Are there other video quiz categories? ▾
Yes — you can also test yourself on AI-generated wildlife footage and sports videos. Each category has different detection challenges and difficulty levels.
