Building artificial people: Endless possibilities and the dark side
summary
A recurrent neural network trained on 14 hours of Obama video synthesizes new lip‑sync that matches arbitrary audio but only the mouth is synthesized while the rest of the face and background come from source clips; a follow‑up method lets a single target video be driven by another sequence to transfer head and facial motion; cycle/recycle GAN variants transfer motion between videos; progressive GANs/StyleGAN generate fully synthetic faces and automatically discover editable attributes like hair color, smile, beard and age.
AI-written summary · about
chapters (6)
- 0:00Start
- 0:18Synthesizing Speech Video from Voice
- 1:20Facial Reenactment (Recycle-GANS)
- 1:51Recycle GANS
- 2:02Generating faces from scratch
- 2:42Use Cases of Face Generative Model
related talks
56mDeep Fakes 2.0 - How Neural Networks are Changing our World
45mGenerative AI for Image and Video Production
13mNavigating Generative AI and Its Risks
58mPushing Deepfakes to the Limit-Fake video calls with AI
37mHarnessing the power of Generative Adversarial Networks (GANS) for supervised learning
51mPlaying Offense and Defense with Deepfakes
15mDrag Makeovers, via AI!
01h 27mMaking Faces: Conditional Generation of Faces using GANs...