WildActor: Unconstrained Identity-Preserving Video Generation

Qin Guo, Tianyu Yang, Xuanhua He, Fei Shen, Yong Zhang, Zhuoliang Kang, Xiaoming Wei, Dan Xu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:38562-38575, 2026.

Abstract

Production-ready human video generation requires digital actors to maintain strictly consistent full-body identities across dynamic shots, viewpoints and motions, a setting that remains challenging for existing methods. Prior methods often suffer from face-centric behavior that neglects body-level consistency, or produce copy-paste artifacts where subjects appear rigid due to pose locking. We present Actor-18M, a large-scale human video dataset designed to capture identity consistency under unconstrained viewpoints and environments. Actor-18M comprises 1.6M videos with 18M corresponding human images, covering both arbitrary views and canonical three-view representations. Leveraging Actor-18M, we propose WildActor, a framework for any-view conditioned human video generation. We introduce an Asymmetric Identity-Preserving Attention mechanism coupled with a Viewpoint-Adaptive Monte Carlo Sampling strategy that iteratively re-weights reference conditions by marginal utility for balanced manifold coverage. Evaluated on the proposed Actor-Bench, WildActor consistently preserves body identity under diverse shot compositions, large viewpoint transitions, and substantial motions, surpassing existing methods in these challenging settings.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-guo26ad, title = {{W}ild{A}ctor: Unconstrained Identity-Preserving Video Generation}, author = {Guo, Qin and Yang, Tianyu and He, Xuanhua and Shen, Fei and Zhang, Yong and Kang, Zhuoliang and Wei, Xiaoming and Xu, Dan}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {38562--38575}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/guo26ad/guo26ad.pdf}, url = {https://proceedings.mlr.press/v306/guo26ad.html}, abstract = {Production-ready human video generation requires digital actors to maintain strictly consistent full-body identities across dynamic shots, viewpoints and motions, a setting that remains challenging for existing methods. Prior methods often suffer from face-centric behavior that neglects body-level consistency, or produce copy-paste artifacts where subjects appear rigid due to pose locking. We present Actor-18M, a large-scale human video dataset designed to capture identity consistency under unconstrained viewpoints and environments. Actor-18M comprises 1.6M videos with 18M corresponding human images, covering both arbitrary views and canonical three-view representations. Leveraging Actor-18M, we propose WildActor, a framework for any-view conditioned human video generation. We introduce an Asymmetric Identity-Preserving Attention mechanism coupled with a Viewpoint-Adaptive Monte Carlo Sampling strategy that iteratively re-weights reference conditions by marginal utility for balanced manifold coverage. Evaluated on the proposed Actor-Bench, WildActor consistently preserves body identity under diverse shot compositions, large viewpoint transitions, and substantial motions, surpassing existing methods in these challenging settings.} }
Endnote
%0 Conference Paper %T WildActor: Unconstrained Identity-Preserving Video Generation %A Qin Guo %A Tianyu Yang %A Xuanhua He %A Fei Shen %A Yong Zhang %A Zhuoliang Kang %A Xiaoming Wei %A Dan Xu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-guo26ad %I PMLR %P 38562--38575 %U https://proceedings.mlr.press/v306/guo26ad.html %V 306 %X Production-ready human video generation requires digital actors to maintain strictly consistent full-body identities across dynamic shots, viewpoints and motions, a setting that remains challenging for existing methods. Prior methods often suffer from face-centric behavior that neglects body-level consistency, or produce copy-paste artifacts where subjects appear rigid due to pose locking. We present Actor-18M, a large-scale human video dataset designed to capture identity consistency under unconstrained viewpoints and environments. Actor-18M comprises 1.6M videos with 18M corresponding human images, covering both arbitrary views and canonical three-view representations. Leveraging Actor-18M, we propose WildActor, a framework for any-view conditioned human video generation. We introduce an Asymmetric Identity-Preserving Attention mechanism coupled with a Viewpoint-Adaptive Monte Carlo Sampling strategy that iteratively re-weights reference conditions by marginal utility for balanced manifold coverage. Evaluated on the proposed Actor-Bench, WildActor consistently preserves body identity under diverse shot compositions, large viewpoint transitions, and substantial motions, surpassing existing methods in these challenging settings.
APA
Guo, Q., Yang, T., He, X., Shen, F., Zhang, Y., Kang, Z., Wei, X. & Xu, D.. (2026). WildActor: Unconstrained Identity-Preserving Video Generation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:38562-38575 Available from https://proceedings.mlr.press/v306/guo26ad.html.

Related Material