Artificial Analysis 榜单第二，SkyReels-V4 宣告 AI 视频进入「全栈统一」阶段

Core Insights - The article highlights the impressive performance of SkyReels-V4 from Kunlun Tiangong, which ranked second in the latest AI video leaderboard by Artificial Analysis, trailing only behind Kuaishou's Kling 3.0 Pro by three ELO points [1][2] - SkyReels-V4's capabilities are distinguished by its unique approach to video generation, which integrates both visual and audio elements, achieving a high level of synchronization and quality [4][5] Performance Ranking - In the global leaderboard, SkyReels-V4 achieved an ELO score of 1090, placing it second overall, while in the historical ranking, it secured the fourth position [2][3] Unique Features - The Text To Video Leaderboard evaluates complete videos with audio, considering both visual quality and audio synchronization, which sets it apart from other models [4] - SkyReels-V4 demonstrates advanced capabilities in motion reference, allowing for seamless character replacement in videos while maintaining the original timing and movements [12][18] Full-Stack Capabilities - The model aims to cover the entire video creation workflow, from generation to editing, all within a single framework, significantly simplifying the creative process [20][34] - It can generate short drama segments with coherent dialogue, background music, and appropriate camera angles, showcasing its ability to understand and implement cinematic language [25][28] Technical Innovations - The underlying technology of SkyReels-V4 includes a unified splicing framework that allows various video tasks to be executed under the same operational model, enhancing efficiency [39][40] - The dual-stream MMDiT architecture enables real-time synchronization of audio and video, ensuring that both elements are generated in harmony [41][44] Industry Implications - The advancements represented by SkyReels-V4 reflect a broader trend in the AI industry towards unifying capabilities across different modalities, which could redefine workflows in content creation [45][46] - The model's ability to perform tasks traditionally requiring multiple specialized tools suggests a potential shift in the industry, particularly in the production of short videos and brand content [46][47]