Context Forcing: Consistent Autoregressive Video Generation with Long Context

Shuo Chen, Cong Wei, Sun Sun, Tiancheng Shen, Ping Nie, Kai Zou, Ge Zhang, Ming-Hsuan Yang, Wenhu Chen
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:17329-17343, 2026.

Abstract

Recent approaches to real-time long video generation typically employ streaming tuning strategies, attempting to train a long-context student using a short-context (memoryless) teacher. In these frameworks, the student performs long rollouts but receives supervision from a teacher limited to short 5-second windows. This structural discrepancy creates a critical student-teacher mismatch: the teacher’s inability to access long-term history prevents it from guiding the student on global temporal dependencies, effectively capping the student’s context length. To resolve this, we propose Context Forcing, a novel framework that trains a long-context student via a long-context teacher. By ensuring the teacher is aware of the full generation history, we eliminate the supervision mismatch, enabling the robust training of models capable of long-term consistency. To make this computationally feasible for extreme durations (e.g., 2 minute), we introduce a context management system that transforms the linearly growing context into a Slow-Fast Memory architecture, significantly reducing visual redundancy. Extensive results demonstrate that our method enables effective context lengths exceeding 20 seconds—$6\text{–}10\times$ longer than state-of-the-art methods like LongLive and Infinite-RoPE. By leveraging this extended context, Context Forcing preserves superior consistency across long durations, surpassing state-of-the-art baselines on various long video evaluation metrics.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-chen26eu, title = {Context Forcing: Consistent Autoregressive Video Generation with Long Context}, author = {Chen, Shuo and Wei, Cong and Sun, Sun and Shen, Tiancheng and Nie, Ping and Zou, Kai and Zhang, Ge and Yang, Ming-Hsuan and Chen, Wenhu}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {17329--17343}, 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/chen26eu/chen26eu.pdf}, url = {https://proceedings.mlr.press/v306/chen26eu.html}, abstract = {Recent approaches to real-time long video generation typically employ streaming tuning strategies, attempting to train a long-context student using a short-context (memoryless) teacher. In these frameworks, the student performs long rollouts but receives supervision from a teacher limited to short 5-second windows. This structural discrepancy creates a critical student-teacher mismatch: the teacher’s inability to access long-term history prevents it from guiding the student on global temporal dependencies, effectively capping the student’s context length. To resolve this, we propose Context Forcing, a novel framework that trains a long-context student via a long-context teacher. By ensuring the teacher is aware of the full generation history, we eliminate the supervision mismatch, enabling the robust training of models capable of long-term consistency. To make this computationally feasible for extreme durations (e.g., 2 minute), we introduce a context management system that transforms the linearly growing context into a Slow-Fast Memory architecture, significantly reducing visual redundancy. Extensive results demonstrate that our method enables effective context lengths exceeding 20 seconds—$6\text{–}10\times$ longer than state-of-the-art methods like LongLive and Infinite-RoPE. By leveraging this extended context, Context Forcing preserves superior consistency across long durations, surpassing state-of-the-art baselines on various long video evaluation metrics.} }
Endnote
%0 Conference Paper %T Context Forcing: Consistent Autoregressive Video Generation with Long Context %A Shuo Chen %A Cong Wei %A Sun Sun %A Tiancheng Shen %A Ping Nie %A Kai Zou %A Ge Zhang %A Ming-Hsuan Yang %A Wenhu Chen %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-chen26eu %I PMLR %P 17329--17343 %U https://proceedings.mlr.press/v306/chen26eu.html %V 306 %X Recent approaches to real-time long video generation typically employ streaming tuning strategies, attempting to train a long-context student using a short-context (memoryless) teacher. In these frameworks, the student performs long rollouts but receives supervision from a teacher limited to short 5-second windows. This structural discrepancy creates a critical student-teacher mismatch: the teacher’s inability to access long-term history prevents it from guiding the student on global temporal dependencies, effectively capping the student’s context length. To resolve this, we propose Context Forcing, a novel framework that trains a long-context student via a long-context teacher. By ensuring the teacher is aware of the full generation history, we eliminate the supervision mismatch, enabling the robust training of models capable of long-term consistency. To make this computationally feasible for extreme durations (e.g., 2 minute), we introduce a context management system that transforms the linearly growing context into a Slow-Fast Memory architecture, significantly reducing visual redundancy. Extensive results demonstrate that our method enables effective context lengths exceeding 20 seconds—$6\text{–}10\times$ longer than state-of-the-art methods like LongLive and Infinite-RoPE. By leveraging this extended context, Context Forcing preserves superior consistency across long durations, surpassing state-of-the-art baselines on various long video evaluation metrics.
APA
Chen, S., Wei, C., Sun, S., Shen, T., Nie, P., Zou, K., Zhang, G., Yang, M. & Chen, W.. (2026). Context Forcing: Consistent Autoregressive Video Generation with Long Context. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:17329-17343 Available from https://proceedings.mlr.press/v306/chen26eu.html.

Related Material