Abstract: To tackle the problem that the existing point cloud-based neural rendering network cannot render high-quality hair with temporal stability, a depth peeling and temporal refine network is presented. Depth peeling method projects point clouds in different layers; fuses the results to adapt to the translucency of the hair; input the trained results into the temporal refine network. This module uses the reprojection of the point cloud between adjacent frames to obtain the dependency relationship between the current frame and the previous frames, and generates the final result of the current frame with temporal stability. The experiment uses high-quality hair datasets generated by ray tracing, and the final results show that compared with the existing methods, the proposed method can obtain better temporal stability and rendering results.