LongTake improves long video generation with sustained dynamics
LongTake: Learning to Sustain Dynamics in Long-Horizon Video Generation
Computer Vision and Pattern Recognition
Summary
Generating long videos that keep moving naturally and look good is hard because models usually focus on short clips during training. The authors propose LongTake, a training method that teaches video models to predict further into the future using longer real videos. This helps the model keep the scene changing smoothly over longer time periods while maintaining image quality. Their approach shows better results than previous methods when creating videos lasting up to a minute.
What this means in practice
- •For video content creators: Generate long continuous video scenes with natural motion for films or virtual environments using improved long-term video prediction techniques.
- •For game developers: Improve simulation of evolving game scenes over extended playtime by sustaining dynamic changes and visual quality in generated video sequences.
Authors
Byoungwoo Park, Jaemoo Choi, Juho Lee, Yongxin Chen
Abstract
World models, game simulators, and long-take video creation require coherent scene evolution and sustained dynamics over extended durations. Autoregressive (AR) video diffusion provides a natural framework for long-horizon generation, yet extended rollouts often become near-static or lose visual quality. We hypothesize that these failures reflect the limited guidance provided by short-video supervision on how ongoing scene dynamics develops over longer durations. This motivates us to introduce LongTake, a two-stage training pipeline built around Long-Horizon Teacher Forcing (TF) on curated real long videos. Long-Horizon TF trains the AR model to predict later frames conditioned on long ground-truth video prefixes, extending direct supervision beyond the short training horizon. This supervision is designed to help the model sustain dynamics and preserve visual quality during long-horizon generation. Our central finding is that this training stage strengthens direct initialization for distribution matching distillation (DMD) under student self-rollout, without the intermediate few-step distillation stage used in standard pipelines. Under the same five-second DMD training setup, our initialization yields substantially higher dynamic degree than short horizon TF initialization on 30-second rollouts at comparable aesthetic quality, and surpasses the evaluated baselines in both measures. Hybrid DMD further reuses this teacher to extend supervision to later frames of the self-rollout while retaining bidirectional joint supervision over the initial window. On long-horizon self-rollouts, LongTake lies on the Pareto front of dynamic degree and aesthetic quality, and Hybrid DMD attains the highest dynamic degree among evaluated methods at both 30s and 60s.