Predicting scene evolution to improve autoregressive video generation
In brief
- Autoregressive models generate videos in successive segments.
- FCP predicts distant segments to enhance physical plausibility.
- The method reduces compute costs without changing inference.
Autoregressive video diffusion models generate videos in successive segments, but their ability to represent scene evolution remains limited. A new method, Future Chunk Prediction (FCP), adds an auxiliary objective that predicts temporally distant segments without using intermediate visual context. According to the announcement, FCP improves the physical plausibility of generated videos while reducing compute costs, without altering autoregressive inference.
This summary gives you the essentials. The full article, its figures and quotes are on Hugging Face Daily Papers.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error