Are you focusing on a (e.g., India, Pakistan)? Should we include technical details on how to spot them? Please share your preferences to tailor the next draft.

At its core, AI-generated media relies on sophisticated machine learning models. The primary engine behind deepfakes is the Generative Adversarial Network (GAN). A GAN consists of two neural networks working in opposition:

Desifakes—AI-generated audio, images, and video that depict South Asian people, languages, and cultural contexts—sit at the intersection of cutting‑edge machine learning and complex sociocultural realities. They raise technical, ethical, political, and cultural questions that deserve sustained, nuanced treatment. Below is a structured, rigorous composition that surveys the phenomenon, explains how it works, outlines harms and opportunities, and proposes concrete interventions for policy, technology, and community resilience.

The "DesiFakes" ecosystem relies on a handful of automated applications and Telegram bots. These tools allow a user to take a single clear photo from a social media profile (Facebook, Instagram, LinkedIn) and map it onto a source video of an adult performer. Within minutes, the AI generates a video where the victim appears to be performing sexual acts.

A GAN operates through two competing neural networks: the generator, which creates the synthetic media, and the discriminator, which evaluates its authenticity. Over millions of iterations, the generator learns to produce hyper-realistic faces, bodies, and voices that seamlessly mimic real human beings.

What used to take weeks on high-end servers can now be done in hours or minutes using cloud computing or consumer-grade GPUs.