Vision
121d agoApple Debuts STARFlow-V Video Generative Model
Apple researchers developed STARFlow-V to challenge the dominance of diffusion models in video generation. This system utilizes normalizing flows to enable end-to-end learning and native likelihood estimation. It handles spatiotemporal complexity more efficiently than previous flow-based attempts.
The Signal
Practitioners gain a robust alternative for causal prediction and high-fidelity video synthesis without relying on iterative denoising.