Alibaba’s Tongyi Wanxiang team has open-sourced Wan2.2-Animate, a model that can animate a single photo into video sequences. It supports motion imitation, where a subject’s actions and expressions are transferred to the photo, and character replacement, where a video character is swapped with the photo subject. Compared with the earlier Animate Anyone model, Wan2.2-Animate improves identity consistency and generation quality, and integrates skeletal and facial signals to better capture body movements and expressions. The model is available on GitHub, Hugging Face, ModelScope, and via API on Alibaba Cloud. [IThome, in Chinese]
Related
