Drag Makeovers, via AI!
summary
StyleGAN encodes faces in a 9000-dimensional latent space trained on ~70k Flickr faces and provides pre-trained models plus a live webcam demo at leebutterman.com. The project finds encodings for queens in and out of drag, interpolates between them, and produces plausible makeovers but requires GPU time (days to weeks at tens of dollars per hour), the encodings converge slowly, and pushing them too far produces visible glitches and dataset bias.
AI-written summary · about
chapters (8)
- 0:00AI-powered drag makeovers
- 1:29StyleGAN face generation
- 3:35Encoding faces and loss
- 6:25Vector space drag
- 9:34Live drag makeover demo
- 10:59Inclusive AI training data
- 12:03Building inclusive AI systems
- 13:31Ethics in AI drag modeling
related talks
37mHarnessing the power of Generative Adversarial Networks (GANS) for supervised learning
56mDeep Fakes 2.0 - How Neural Networks are Changing our World
03mBuilding artificial people: Endless possibilities and the dark side
27m"AI perceptions of gender"
58mStyle Transfer AI - How Neural Networks Generate Art
06mGAN-based Style Transformation to Improve Gesture-recognition Accuracy
58mPushing Deepfakes to the Limit-Fake video calls with AI
36mArt in the Age of Algorithms