阿里AI四连发，宣布开源电影级视频模型Wan2.2

Core Insights - Alibaba has launched a new open-source model called Wan2.2, which enhances video generation capabilities by incorporating cinematic aesthetics such as light, color, and camera language [2][3] - The model allows users to generate 5 seconds of high-definition video in a single instance and aims to improve video creation efficiency in the future [2] - Wan2.2 includes three models: text-to-video (Wan2.2-T2V-A14B), image-to-video (Wan2.2-I2V-A14B), and unified video generation (Wan2.2-TI2V-5B), all utilizing a MoE architecture [2] Technical Features - The total parameter count for the models is 27 billion, with 14 billion active parameters, designed to optimize resource consumption by approximately 50% [2] - The models are structured with high-noise expert models for overall layout and low-noise expert models for detail enhancement, addressing the challenges of long token processing in video generation [2] - Significant improvements have been made in complex motion generation, character interaction, and aesthetic expression [2] Aesthetic Control - Wan2.2 introduces a cinematic aesthetic control system that can generate professional-level visuals based on user-defined keywords such as "dusk," "soft light," and "warm tones" [3] - The model can produce various visual effects, including a golden sunset or a sci-fi scene, by combining different aesthetic parameters [3]