The method includes acquiring first feature data of a physical object in real time, where the physical object includes a torso, a main arm, an elbow joint, and a forearm, the first feature data is configured for representing rotation angles of multiple parts of the forearm around an axial direction, and the rotation angles of the multiple parts of the forearm around the axial direction are positively correlated with distances from the multiple parts of the forearm to the elbow joint of the physical object; controlling rotation angles of multiple parts of the forearm skin of a virtual model around an axis direction based on the first feature data acquired in real time, where the rotation angles of the multiple parts of the forearm skin around the axial direction are positively correlated with distances from the multiple parts of the forearm skin to the elbow joint of the virtual model; and generating the image frame of the live video according to the virtual model and the rotation angles of the multiple parts of the forearm skin of the virtual model around the axis direction.