This is a synthetic dataset created from Animatable Gaussian Avatars. For each identity, we randomly select a facial expression and render 3D Gaussians along a camera trajectory, obtaining a video stacked with multiple frames. The pose and expression stay static during the rendering process.