Wan-AI/Wan2.2-T2V-A14B

Wan2.2-T2V-A14B is the industry's first open-source video generation model, released by Alibaba, that adopts a Mixture of Experts (MoE) architecture. This model focuses on text-to-video generation tasks and can produce videos lasting 5 seconds with resolutions of either 480P or 720P. By integrating the MoE architecture, the model expands its overall capacity while keeping inference costs nearly unchanged. It features a high-noise expert responsible for handling the initial-stage layout and a low-noise expert that refines fine details in the later stages of video creation. Additionally, Wan2.2 incorporates carefully curated aesthetic data, meticulously annotated across dimensions such as lighting, composition, and color, enabling more precise and controllable generation of cinematic-style outputs. Compared to its predecessor, this model was trained on a significantly larger dataset, greatly enhancing its generalization capabilities in areas like motion, semantics, and aesthetics, allowing it to better manage complex dynamic effects.