Level Up Your Game Dev Journey

Face Anything: A New State-of-the-Art 4D Facial Capture Model

Researchers have introduced “Face Anything,” a transformer-based model for high-quality 4D facial reconstruction and tracking. The system maps every pixel to canonical facial coordinates while jointly predicting depth, turning complex, dynamic faces into a stable canonical reconstruction problem. Benchmarks show roughly 3× lower correspondence error, faster inference, and 16% better depth accuracy than previous dynamic...