Enhancing Video Generation Consistency with Additional Perturbation and Adversarial Training

A novel architecture, APLA, that builds upon pre-trained diffusion models to enhance the consistency between video frames by learning the correlation information between input frames. APLA employs a fusion of diffusion models and adversarial training to improve the quality and consistency of generated videos.