Vision
120d agoApple Debuts STARFlow-V Video Generator
Apple developed STARFlow-V to challenge the dominance of diffusion models in video generation. This new system utilizes normalizing flows to enable end-to-end learning and native likelihood estimation. It handles spatiotemporal complexity more efficiently than traditional iterative methods.
The Signal
Researchers can now implement robust causal prediction without the computational overhead typical of diffusion-based architectures.