Deepfakes are no longer just the stuff of sci-fi thrillers—they’re here, and they’re deceptively good. From celebrity endorsements to real-time impersonations, deepfake technology has advanced to the point where spotting one isn’t as easy as it used to be.
In this post, I’ll share insights from my own testing and experimenting with current deepfake creation technologies. You’ll get a behind-the-scenes look at how they’re made and learn what to watch out for so you can stay ahead of the game.
Understanding the Threat
Deepfakes are synthetic media where someone’s face, voice, or both are convincingly replaced or manipulated. They’re often used for scams, misinformation, and fraud. For instance, scammers have used deepfakes to impersonate executives in video calls or create fake celebrity endorsements for products.
The technology behind deepfakes, like DeepFaceLab/DeepFaceLive or Deep Live Cam, has made creating these fakes more accessible than ever. The easy access to these tools enables creative and educational uses, but it also lowers the barriers for malicious purposes. Cybercriminals and scammers often have the motivation and time to research and master these tools, while Red Teamers and Security Awareness professionals are frequently stretched thin with limited time and resources.
Because of that, I recently created a series of YouTube videos helping Red Teamers and Security Awareness leaders get up-to-speed on the technology, techniques, and detection methods. As of today, this series consists of three videos. I think of the series as: The Defenders Guide to Understanding, Creating, and Detecting Deepfakes. The series includes:
- Inside a celebrity deepfake: How I Made Taylor Swift 'Endorse' My Book
- How to create real-time deepfakes (a.k.a. I became Taylor Swift...for Science!)
- Deepfake SECRETS EXPOSED: Outsmart AI Deception with These Tricks!
The latest in the series (Deepfake SECRETS EXPOSED: Outsmart AI Deception with These Tricks!) is all about some of the oddities and tells that exist in current deepfakes… and that’s what I’d like to spend a bit of time covering in this blog post.
Common Red Flags in Deepfakes
Keep in mind that the technology is constantly improving. Absence of a tell does not mean that something is not a deepfake. That being said, here are a few things to look out for that are indicative of current issues with today’s most commonly used deepfake creation programs. I’ve illustrated many of these with screengrabs from the video.
-
Visual Oddities
Side Adhesion Issues: Many deepfakes struggle with maintaining facial integrity when the subject turns their head to the side. Watch for distortions or unnatural blending around the edges of the face.
Figure 1: Notice the side profile adhesion issues. These will become less common over time, but still are still a problem with some deepfake software.
Edge Artifacts: Look for digital interference or halo effects around the subject, particularly near text or signs.
Figure 2: Me as Nicolas Cage pointing out the artifacts on the side behind me. Notice how most everything else is clear and in focus.
Masking Problems: When a hand or object passes in front of the face, the underlying image might ghost or reveal the mask.
Figure 3: Notice how the Mask wants to pick through even when you cover the face.
-
Missing or Inconsistent Features: Depending on the software used, sometimes features like birthmarks, scars, makeup, or other aspects of the face can change or become inconsistent.
No Tongue: Many deepfakes fail to render a tongue when the subject opens their mouth.
Figure 4: Many deepfakes fail to render a tongue when the subject opens their mouth
Figure 5: Another example of the tongue issue.
Beard or Makeup Mismatch: Facial hair or heavy makeup can bleed through or disappear, especially in single-image deepfake software. This is because single-image, deep, big software focuses primarily on mapping the topology of the face and creating a 3D model of that to be integrated with the face of the person who is going to wear the mask.
Figure 6: Here is an extreme example. I am attempting to do a deep fake of Joaquin Phoenix's portrayal of the Joker... instead, I just get Joaquin Phoenix.
-
Motion Anomalies
-
Lagging Frames: Deepfakes often have a lower framerate than the rest of the video, resulting in choppy motion.
-
Lighting Adjustments: Cameras compensating for poor lighting can amplify deepfake artifacts.
-
Voice and Lip Sync Discrepancies
-
Lip sync issues are still common in both real-time and pre-recorded deepfakes. Poor coordination between speech, facial expressions, and gestures is a major giveaway.
Real-Time Detection Tips
Video calls present unique opportunities for deepfake detection because you can interact with and challenge the subject in real-time. While some artifacts might be dismissed as normal video conferencing issues, the combination of multiple red flags should raise suspicion. Here are key tests to try:
- Side Profile Test: Ask the person to turn their head to the side. Deepfake systems often distort or fail to maintain integrity at extreme angles.
- Hand Interaction Test: Ask them to place a hand or a finger in front of their face. Current deepfake technology struggles with occlusion, often creating ghosting effects or unnatural masking.
- Tongue Test: Request they stick out their tongue. Many deepfakes can’t render tongues or show obvious artifacts.
- Mouth-Audio Sync: Pay attention to consistent patterns of misalignment. Natural lag tends to be intermittent, while deepfake desynchronization remains a constant issue.
Critical Questions to Ask
In a world where we can’t trust our eyes and ears alone, here’s a new framework for assessing content:
Figure 7: Four Critical Questions to Ask Yourself
- Why am I seeing this? What’s the context, and why was this content put in front of me?
- Who created this? Can you verify the source?
- What’s the intent? Is someone trying to sell me something, manipulate my beliefs, or extract sensitive information?
- What emotions are they tapping into? Scammers often play on urgency, fear, or excitement to cloud judgment.
Practical Defense Strategies
Implementing effective defense strategies requires a combination of technical awareness and practical actions:
For Video Calls
- Establish verification protocols for sensitive communications
- Use multi-factor authentication and processes beyond visual confirmation
- Request specific, unpredictable movements or interactions
- Maintain secondary communication channels for verification
Conclusion
As deepfake technology continues to evolve, traditional visual detection methods will become increasingly unreliable. The future of digital media authentication lies in developing strong critical thinking skills and maintaining healthy skepticism about digital content. Rather than trying to spot every deepfake, focus on understanding context, motivation, and potential manipulation attempts.
Building these skills isn’t about becoming paranoid; it’s about developing digital resilience. By understanding deepfake capabilities and limitations, we can better protect ourselves and others from digital deception while still enjoying the benefits of online communication.
If you want more information about how to live in a world filled with deepfakes, disinformation, and AI-generated deceptions, check out my newest book, FAIK, as well as my new YouTube channel.
Final Recommendations
- Stay informed about emerging deepfake technologies
- Share knowledge with your community
- Trust your instincts when something feels off
- Verify important information through multiple channels
Learn about how AIDA can significantly reduce human risk, streamline security operations and support a strong security culture in your organization.
Learn More About AIDA