Researchers propose Factor-Aware Predictive Learning for cinematic video generation

The paper argues that multimodal large language models provide useful priors but their representations are not automatically actionable for video generation; a Video Diffusion Transformer may bypass them through native conditioning pathways.