FaceCam: Portrait Video Camera Control via Scale-Aware Conditioning

Summary

FaceCam is a system introduced by Weijie Lyu, Ming-Hsuan Yang, and Zhixin Shu that generates portrait videos with customizable camera trajectories from monocular human portrait video input. Recent camera control approaches built on large video-generation models have made promising progress, but they often produce geometric distortions and visual artifacts on portrait videos because of scale-ambiguous camera representations or 3D reconstruction errors. To address these limitations, the authors propose a face-tailored scale-aware representation for camera transformation, which conditions the video generation process on explicitly scale-aware signals rather than ambiguous or reconstructed camera parameters. The work is categorized under computer vision (cs.CV) and is available on arXiv as paper 2603.05506. This post is an automated arXiv digest shared on zhichai.net, linking to the original abstract page for readers who want the full details, method descriptions, and experimental results.