Vision
119d agoApple Debuts STARFlow-V Video Generator
Apple researchers developed STARFlow-V, a video generator utilizing normalizing flows instead of the industry-standard diffusion models. This architecture enables end-to-end learning and native likelihood estimation for complex spatiotemporal data. It offers a robust alternative for causal prediction in video synthesis.
The Signal
Practitioners gain a more computationally efficient path toward likelihood-based generative modeling for high-dimensional video streams.