Animations are extremely important to augment the believability and effectiveness of non-player characters (NPCs) for game users. Our session describes a novel approach, relying on deep learning technologies, to automatically synthesize high-quality and life-like full-body animations for talking NPCs. The animations involve lip and chin, upper facial expression (eyebrows, upper and bottom eyelids, and eyeballs), head rotation, torso and hand gestures, and legs and feet. The synthesized animations can reflect speech prosody, express the emotional state expressed by speech and correspond to specific personalities and professions of NPCs. Without any manual intervention, it takes only less than 500ms to compute full-body animation trajectories for an utterance which lasts about 5 to 20 seconds. The automatic animation generator allows animators or artists to be released from manually creating animations and processing motion capture data.