[NeurIPS 2024] Diffusion4D: Fast Spatial-temporal Consistent 4D Generation via Video Diffusion Models