Next Block Prediction: Video Generation via Semi-Autoregressive Modeling

TL;DR

This paper introduces Next-Block Prediction, a semi-autoregressive framework for video generation that uses block-based bidirectional attention to enable parallel token prediction, resulting in faster inference and improved quality.

Contribution

The paper proposes a novel semi-autoregressive model for video generation that reduces inference steps and enhances quality through block-based bidirectional attention.

Findings

Achieves state-of-the-art FVD scores of 103.3 on UCF101 and 25.5 on K600.

Generates 8.89 frames per second at 128x128 resolution, 11x faster than previous models.

Scaling the model from 700M to 3B parameters improves FVD scores significantly.

Abstract

Next-Token Prediction (NTP) is a de facto approach for autoregressive (AR) video generation, but it suffers from suboptimal unidirectional dependencies and slow inference speed. In this work, we propose a semi-autoregressive (semi-AR) framework, called Next-Block Prediction (NBP), for video generation. By uniformly decomposing video content into equal-sized blocks (e.g., rows or frames), we shift the generation unit from individual tokens to blocks, allowing each token in the current block to simultaneously predict the corresponding token in the next block. Unlike traditional AR modeling, our framework employs bidirectional attention within each block, enabling tokens to capture more robust spatial dependencies. By predicting multiple tokens in parallel, NBP models significantly reduce the number of generation steps, leading to faster and more efficient inference. Our model achieves FVD scores of 103.3 on UCF101 and 25.5 on K600, outperforming the vanilla NTP model by an average of 4.4. Furthermore, thanks to the reduced number of inference steps, the NBP model generates 8.89 frames (128x128 resolution) per second, achieving an 11x speedup. We also explored model scales ranging from 700M to 3B parameters, observing significant improvements in generation quality, with FVD scores dropping from 103.3 to 55.3 on UCF101 and from 25.5 to 19.5 on K600, demonstrating the scalability of our approach.