dailyai.report

Apple researchers developed STARFlow-V, a video generator utilizing normalizing flows instead of the industry-standard diffusion process. This architecture enables end-to-end learning and native likelihood estimation. It challenges the current dominance of diffusion models by reducing computational overhead during inference. Practitioners gain a robust alternative for causal prediction in complex spatiotemporal data.