How I built the world’s most efficient deepfake detector with $100
summary
Claims a 100% reliable detector for images from ThisPersonDoesNotExist.com and a method to recover the exact generation timestamp by continuously scraping the site (it produces ~72,000 images/day ≈ 1/sec and often serves identical cached images globally). Implementation runs on a $100-ish Jetson Nano, computes FaceNet visual hashes for huge storage reduction (~200×) and indexes vectors with OpenSearch KNN, enabling sub-second similarity lookups and robust matches despite social-media recompression, flips or small edits.
AI-written summary · about
chapters (8)
- 0:00The deepfake problem
- 2:51Understanding neural networks
- 5:52How GANs function
- 14:13Detecting AI-generated faces
- 19:55Building a detection system
- 30:57System architecture overview
- 33:47System demo and results
- 40:04Future implications
related talks
47mHow I designed the most efficient deepfake detector in the world with $100
35mHow I Built the World's Most Powerful Deepfake Detector for €100
56mDeep Fakes 2.0 - How Neural Networks are Changing our World
27mIs now the time to solve the deepfake threat?
51mPlaying Offense and Defense with Deepfakes
30mAI Gone Rogue: Exterminating Deep Fakes Before They Cause Menace
32mHeart to Heart and Eye to Eye with Deep Fakes
01h 04mPushing Deepfakes to The Limit - Fake Video Calls With AI