Empirical and Comparative Study of Long-Sequence Video Consistency in AIGC

Abstract

With the rise of generative AI, AI-based video synthesis has emerged as a transformative tool in film, advertising, and new media. However, complex scenes continue to face challenges, such as temporal discontinuity, lack of physical consistency, and style shifts. This study conducts a comparative analysis of JiMeng, Vidu, and Keling AI across six scenarios: forest/animals, city/street, indoor/people, beach/nature, sci-fi/city, and product/exhibition. Using unified prompts and a standardized frame-continuity strategy, 5-s videos (16:9) were generated under default settings. Results show that JiMeng performs best in urban, sci-fi, and product scenes; Keling excels in natural environments; and Vidu stands out in indoor character expressions. This study proposes a platform evaluation paradigm and highlights scenario-specific strengths, providing essential technical guidance for creative applications in the AI-driven media landscape.