The technology behind Wan-Animate is grounded in the Wan model, adapted specifically for animation purposes through a modified input paradigm. This allows the system to handle reference conditions and generation regions in a unified symbolic representation. The framework uses spatially-aligned skeleton signals to control body motions and extracts implicit facial features to reenact expressions with a high degree of control and expressiveness, enabling the creation of detailed and natural-looking character animations.
To ensure that replaced characters blend perfectly within their new environments, Wan-Animate incorporates an auxiliary Relighting LoRA module. This specialized component maintains the consistency of the character's appearance while applying environment-specific lighting and color tones, enhancing the overall realism of the animation. The creators emphasize state-of-the-art performance and commitment to openness by providing the model weights and source code to the public for further development and use.
