WELCOME
DEPARTMENT OF
COMPUTER ENGINEERING
Seminar Presentation – 12-09-2025
DeepFake Detection Technologies
Presented By
SHINAN.K.T
OUTLINE
• Introduction
• Applications of Deepfakes
• Types of Deepfakes
• Popular Tools & Creation Process
• Dangers & Threats Overview
• Traditional Detection Methods
• Multimodal Detection Systems
• Real-time Detection Challenges
• Ethical Considerations
• Challenges in Detection
• Impact on National Security
• Case Study
• Future Trends in Detection
• Conclusion
What are Deepfakes?
• Deepfakes are synthetic media created using
AI, primarily deep learning techniques like
GANs.
Introduction to Deepfakes
• Deepfakes use artificial intelligence to
manipulate or generate visual and audio
content.
• They are increasingly used for both positive
and harmful purposes.
Applications of Deepfakes
• Entertainment and media dubbing
• Education and historical re-creation
• Virtual influencers
Types of Deepfakes
• Face swapping
• Voice cloning
• Lip-sync deepfakes
• Text-to-video generation
Popular Deepfake Tools
• DeepFaceLab
• FaceSwap
• FakeApp
How Deepfakes Are Made
• Process involves data collection, model
training and rendering.
• Training on large datasets of images and
voices improves realism.
• Advanced tools: DeepFaceLab, FaceSwap.
Transformer-based Methods
• Text-to-image transformers
• Integration in multimedia fakes
Dangers & Threats Overview
• Misinformation
• Identity theft
• Financial fraud
• Political influence
Misinformation & Fake News
• Deepfake news articles
• Impact on public opinion
Financial Fraud
• Voice cloning in banking
• CEO fraud cases
Deepfake in Cybersecurity
• Voice impersonation in banking
• Hacking facial recognition systems
Political Manipulation
• Election interference
• Fake campaign speeches
Traditional Detection Methods
• Visual artifact analysis
• Audio inconsistencies
• Metadata verification
Biometric-based Approaches
• Eye blink rate analysis
• Lip-sync detection
Video Deepfake Detection
• Frame-by-frame analysis helps detect irregular
facial behavior or lighting.
Detection with Forensics
• Inconsistencies in face lighting, blinking
patterns.
• Texture artifacts in skin and hair.
• Mismatch between audio and lip movements.
Multimodal Detection Systems
• Combining audio, video, text
• Case studies
On-Device Detection
• Edge computing solutions
• Lightweight models
• Mobile SDKs
Real-time Detection Challenges
• Latency constraints
• GPU and CPU load
Cloud-based Detection Services
• API offerings (Microsoft, AWS)
• Cost considerations
Ethical Considerations
• Consent and privacy
• Responsible AI principles
Watermarking Deepfakes
• Invisible signatures
Detection Challenges
• High-quality fakes, evolving AI, and lack of
data make detection difficult.
Impact on National Security
• Threat to democracy, information warfare,
and cyber defense.
Public Awareness & Education
• Digital literacy programs
• Spot-the-fake campaigns
Case Study: Obama Speech
• Overview of fake Obama video
• Detection methods used
• Lessons learned
Future Trends in Detection
• Hardware-based authentications
• AI model advances
• Cross-domain verification
Key Takeaways
• Deepfakes: dual-use tech
• Detection requires multi-faceted approach
• Collaboration of tech, law, policy
Conclusion
• deepfakes represent a dual-use technology
with significant potential for both innovation
and harm
• Detection remains a complex challenge due to
evolving Al and high-quality fakes
References
• 1. https://www.media.mit.edu
• 2. https://github.com/facebook/DFDC
• 3. https://sensity.ai
Thank you
