TokensGen: Harnessing Condensed Tokens for Long Video Generation

1S-Lab, Nanyang Technological University, 2SenseTime Research,
3Wangxuan Institute of Computer Technology, Peking University

ICCV 2025

Abstract

Generating consistent long videos is a complex challenge: while diffusion-based generative models generate visually impressive short clips, extending them to longer durations often leads to memory bottlenecks and long-term inconsistency. In this paper, we propose TokensGen, a novel two-stage framework that leverages condensed tokens to address these issues. Our method decomposes long video generation into three core tasks: (1) inner-clip semantic control, (2) long-term consistency control, and (3) inter-clip smooth transition.

First, we train To2V (Token-to-Video), a short video diffusion model guided by text and video tokens, with a Video Tokenizer that condenses short clips into semantically rich tokens. Second, we introduce T2To (Text-to-Token), a video token diffusion transformer that generates all tokens at once, ensuring global consistency across clips. Finally, during inference, an adaptive FIFO-Diffusion strategy seamlessly connects adjacent clips, reducing boundary artifacts and enhancing smooth transitions.

Experimental results demonstrate that our approach significantly enhances long-term temporal and content coherence without incurring prohibitive computational overhead. By leveraging condensed tokens and pre-trained short video models, our method provides a scalable, modular solution for long video generation, opening new possibilities for storytelling, cinematic production, and immersive simulations.

Visual Results

Generation

Editing

Comparisons


Ablation Study


BibTeX

@misc{ouyang2025tokensgenharnessingcondensedtokens,
      title={TokensGen: Harnessing Condensed Tokens for Long Video Generation}, 
      author={Wenqi Ouyang and Zeqi Xiao and Danni Yang and Yifan Zhou and Shuai Yang and Lei Yang and Jianlou Si and Xingang Pan},
      year={2025},
      eprint={2507.15728},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.15728}, 
}