Deepfake Verification Lab
Learn how GANs create convincing fake images, spot the telltale artefacts, play "real or fake", and discover why critical thinking about online media has never been more important!
How AI Creates Deepfakes
Generator
A neural network that creates fake images, starting from random noise and learning to make them look increasingly realistic.
Discriminator
A second neural network that tries to distinguish real images from fake ones — the "detective" that pushes the generator to improve.
Arms Race
Generator and discriminator compete. Generator gets better at faking; discriminator gets better at detecting. Both improve together.
The Problem
Modern GANs produce images indistinguishable to the human eye — fuelling misinformation and fake political content worldwide.
Step 1 — How GANs Work
🤖 GAN Training Status
1. Generator creates fake image
2. Discriminator sees real + fake
3. Discriminator: "Real or fake?"
4. Both networks update their weights
5. Repeat thousands of times
Eventually the fakes are so good the discriminator can only guess randomly — 50% accuracy.
Step 2 — Real or Fake?
Step 3 — Deepfake Artefact Analyser
Step 4 — Deepfake Ethics Scenarios
1. Look for ear asymmetry — GANs often generate mismatched ears
2. Check hair edges — real hair is complex; AI hair often blurs
3. Look at teeth — GANs struggle with teeth detail
4. Check the background — AI backgrounds are often unrealistically smooth
5. Look for facial edge blur — where the face meets hair/background
6. When in doubt — verify the source, use reverse image search
Deepfake Detector Badge!
You mastered GANs, artefact detection, and AI media ethics!
Optional. Stays on this device only — not sent to WhizzStep.
Key Concepts Mastered
🔄 Adversarial Training
Two networks compete: Generator creates fakes, Discriminator detects them. Both improve through competition until fakes are indistinguishable.
🔬 The Tell-tale Signs
GAN artefacts: ear asymmetry, hair edge blurring, teeth irregularities, background smoothness, and facial boundary inconsistencies.
🔁 GAN Failure
When the generator learns to produce only a few types of outputs — it "collapses" into repetitive patterns. A common training failure.
⚖️ Digital Rights
Creating deepfakes of real people without consent is illegal in many countries — including India, where IT Act amendments address synthetic media.
🏃 Always Catching Up
As detection improves, generators improve to fool the detector. This is a genuine adversarial arms race with no clear winner.
🧠 Critical Thinking
The best defence isn't technology — it's humans who question viral media, verify sources, and understand AI limitations.
About this lab
Learning objective: Practise source, context, provenance and uncertainty checks on sample media.
What this simplifies: Visual clues alone do not prove that media are authentic or synthetic; this activity teaches verification habits, not detection certainty.
Privacy: No learner input leaves the device.
Teacher prompt: Ask the class why this simulation might mislead someone who takes it too literally.
Reflect: What is one thing this activity showed you that you did not expect?