Looking Ahead: Modeling Future Evolution in Autoregressive Video Generation
Paper · https://arxiv.org/abs/2610.29989
Autoregressive video diffusion models generate videos chunk by chunk, yet producing plausible continuations does not guarantee that they capture how scenes evolve. Under teacher forcing, strong temporal redundancy may allow local visual continuity to serve as a shortcut for current-chunk denoising, reducing the pressure to model visual evolution. We introduce Future Chunk Prediction (FCP), an auxiliary objective jointly optimized with teacher-forced denoising that predicts temporally separated chunks without access to the intervening visual context. Drawing supervision directly from training videos, FCP requires no additional pretrained visual encoder and leaves autoregressive inference unchanged. Our method achieves superior physical plausibility over standard teacher-forced autoregressive training with a lower overall training compute budget, highlighting the effectiveness and efficiency of future-chunk supervision for modeling visual evolution. Intriguingly, our visualizations suggest that FCP fosters a more structured representation space, which may be valuable for building autoregressive world models.
First seen by Oriel on .
AI summary
The paper introduces Future Chunk Prediction (FCP), an auxiliary training objective that improves physical plausibility in autoregressive video generation by supervising predictions of temporally separated video chunks.
- FCP is jointly optimized with teacher-forced denoising and predicts future chunks without access to intervening visual context.
- The method draws supervision directly from training videos and requires no additional pretrained visual encoder.
- The authors report better physical plausibility than standard teacher-forced autoregressive training with a lower overall training compute budget.
- FCP does not change autoregressive inference, and visualizations suggest it may produce a more structured representation space for world models.
#video generation #diffusion models #world models
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hugging Face Papers: 173, latest observation ; +88 over observed period since
Comments
Comment count reported by the source.
- Hugging Face Papers: 2, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Papers | 2610.29989 | https://arxiv.org/abs/2610.29989 Read discussion |