Samsung Labs researchers are working on MegaPortraits, a system that uses AI-powered neural networks to generate deepfakes from a single photo. Unlike other deepfakes where the subject needed to look similar to the target, the team here advanced neural head avatar technology to the megapixel resolution while focusing on the particularly challenging task of cross-driving synthesis.
This means that the driving image, or subject, can look substantially different from the animated source image. To accomplish this, the AI is trained in two stages, starting by sampling two random frames from a dataset at each step: the source frame and driver frame. Facebook AI can reverse engineer deepfakes, so we don’t recommend using these as your profile photo on social media, or at least on any of Meta’s properties.
- Powerful Productivity: 11th Generation Intel Core i3-1115G4 Dual Core processor delivers unmatched speed and intelligence, enabling impressive...
- Visibly Stunning: Experience sharp details and crisp colors on the 15.6" Full HD IPS display with 82.58% screen-to-body, 16:9 aspect ratio and narrow...
- Ergonomic Typing: Ergonomically-designed hinge lifts the keyboard for comfortable typing, improved cooling, and a better sound experience
We show how a trained high-resolution neural avatar model can be distilled into a lightweight student model which runs in real-time and locks the identities of neural avatars to several dozens of pre-defined source images. Real-time operation and identity lock are essential for many practical applications head avatar systems,” said the researchers.
