On September 2, news emerged that Daxiao Robotics, in partnership with prestigious institutions including The Chinese University of Hong Kong and Nanyang Technological University, has released ACE-Ego-Hand as an open-source model. This innovative model is designed to reconstruct 3D hand shapes from first-person perspective videos and has been trained on an extensive dataset of approximately 5,000 hours. By transforming video diffusion models into deterministic geometric encoders, ACE-Ego-Hand streamlines the process, replacing the traditional multi-step denoising approach with a single feedforward pass. This optimization leads to a remarkable 33-fold increase in inference speed. The technology allows for the direct conversion of human first-person operational videos into structured 3D trajectories, which can then be seamlessly transferred to dexterous robotic hands. This breakthrough significantly eases the data collection process for embodied AI imitation learning, paving the way for more efficient and accessible robotic training methods.