Deepfakes & Synthetic Media
When the Evidence Lies
6 min read
When the Evidence Lies
For centuries, seeing was believing. Deepfakes ended that era.
Deepfakes use AI to convincingly swap faces in video, clone voices from seconds of audio, and generate entirely synthetic but realistic media. The technology that creates them is the same diffusion and generative AI that makes creative tools powerful. In India, deepfake videos of politicians and celebrities have already caused real harm. Knowing they exist — and how to spot them — is now basic digital literacy.
In Plain English
Deepfakes are AI-generated synthetic media — videos, audio, or images — that realistically portray people doing or saying things they never did. They're created using the same generative AI techniques behind image generation. Detection is increasingly difficult and is an active research area.
The Technical Picture
Deepfake video generation uses GANs (face-swap) or diffusion models (full synthesis). Voice cloning uses neural TTS models trained on minutes of audio. Detection approaches include forensic artefact analysis, physiological signal detection (inconsistent blinking, unnatural blood flow patterns in video), and provenance watermarking (C2PA standard).
Real-World Examples
- A deepfake of a Bollywood celebrity endorsing a financial scam circulated on WhatsApp in 2024
- Voice cloning used in CEO fraud calls — convincing audio of executives authorising wire transfers
- C2PA content credentials (supported by Adobe, Microsoft, Google) digitally sign authentic media at capture
Deepfakes make synthetic media indistinguishable from real — verification of sources is now a critical skill.