RIFLEx: A Free Lunch for Length Extrapolation in Video Diffusion Transformers

Min Zhao, Guande He, Yixiao Chen, Hongzhou Zhu, Chongxuan Li, Jun Zhu
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:77539-77557, 2025.

Abstract

Recent advancements in video generation have enabled models to synthesize high-quality, minute-long videos. However, generating even longer videos with temporal coherence remains a major challenge and existing length extrapolation methods lead to temporal repetition or motion deceleration. In this work, we systematically analyze the role of frequency components in positional embeddings and identify an intrinsic frequency that primarily governs extrapolation behavior. Based on this insight, we propose RIFLEx, a minimal yet effective approach that reduces the intrinsic frequency to suppress repetition while preserving motion consistency, without requiring any additional modifications. RIFLEx offers a true free lunch—achieving high-quality $2\times$ extrapolation on state-of-the-art video diffusion transformers in a completely training-free manner. Moreover, it enhances quality and enables $3\times$ extrapolation by minimal fine-tuning without long videos.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-zhao25m, title = {{RIFLE}x: A Free Lunch for Length Extrapolation in Video Diffusion Transformers}, author = {Zhao, Min and He, Guande and Chen, Yixiao and Zhu, Hongzhou and Li, Chongxuan and Zhu, Jun}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {77539--77557}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/zhao25m/zhao25m.pdf}, url = {https://proceedings.mlr.press/v267/zhao25m.html}, abstract = {Recent advancements in video generation have enabled models to synthesize high-quality, minute-long videos. However, generating even longer videos with temporal coherence remains a major challenge and existing length extrapolation methods lead to temporal repetition or motion deceleration. In this work, we systematically analyze the role of frequency components in positional embeddings and identify an intrinsic frequency that primarily governs extrapolation behavior. Based on this insight, we propose RIFLEx, a minimal yet effective approach that reduces the intrinsic frequency to suppress repetition while preserving motion consistency, without requiring any additional modifications. RIFLEx offers a true free lunch—achieving high-quality $2\times$ extrapolation on state-of-the-art video diffusion transformers in a completely training-free manner. Moreover, it enhances quality and enables $3\times$ extrapolation by minimal fine-tuning without long videos.} }
Endnote
%0 Conference Paper %T RIFLEx: A Free Lunch for Length Extrapolation in Video Diffusion Transformers %A Min Zhao %A Guande He %A Yixiao Chen %A Hongzhou Zhu %A Chongxuan Li %A Jun Zhu %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-zhao25m %I PMLR %P 77539--77557 %U https://proceedings.mlr.press/v267/zhao25m.html %V 267 %X Recent advancements in video generation have enabled models to synthesize high-quality, minute-long videos. However, generating even longer videos with temporal coherence remains a major challenge and existing length extrapolation methods lead to temporal repetition or motion deceleration. In this work, we systematically analyze the role of frequency components in positional embeddings and identify an intrinsic frequency that primarily governs extrapolation behavior. Based on this insight, we propose RIFLEx, a minimal yet effective approach that reduces the intrinsic frequency to suppress repetition while preserving motion consistency, without requiring any additional modifications. RIFLEx offers a true free lunch—achieving high-quality $2\times$ extrapolation on state-of-the-art video diffusion transformers in a completely training-free manner. Moreover, it enhances quality and enables $3\times$ extrapolation by minimal fine-tuning without long videos.
APA
Zhao, M., He, G., Chen, Y., Zhu, H., Li, C. & Zhu, J.. (2025). RIFLEx: A Free Lunch for Length Extrapolation in Video Diffusion Transformers. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:77539-77557 Available from https://proceedings.mlr.press/v267/zhao25m.html.

Related Material