Next Forcing: Causal World Modeling with Multi-Chunk Prediction

2026-06-09

AI Summary

Existing autoregressive video generation methods face bottlenecks in training convergence speed, high frame-rate modeling accuracy, and inference efficiency. This work proposes Next Forcing, a Multi-Chunk Prediction (MCP) framework that introduces multi-token prediction into video generation for the first time. By employing a lightweight auxiliary module to predict multiple future video chunks in parallel, Next Forcing constructs a causally chained structure with cross-temporal depth and fuses multi-layer features from the main model to enable multi-scale future prediction. The approach substantially improves training convergence, generation fidelity, and inference speed: it achieves a 93.1% performance gain over LingBot-VA after only 5k training steps at 50 fps, converging 2.3× faster; attains state-of-the-art results on RoboTwin (94.1/93.5%), with 2× faster inference; and significantly enhances physical consistency on PhyWorld, reducing the pretraining FVD by over 50%.

Abstract

Autoregressive video generation has emerged as a powerful paradigm for World Action Models (WAMs). However, existing approaches suffer from slow training convergence and limited converged accuracy, particularly at high frame rates, as the training supervision is confined to the current chunk without explicit signals about future dynamics; they also suffer from slow inference due to iterative video denoising. In this paper, we present Next Forcing, a multi-chunk prediction (MCP) framework for causal world modeling that enables faster training, higher accuracy, and accelerated inference. Inspired by multi-token prediction in large language models, Next Forcing introduces an MCP training objective that augments the main model with lightweight auxiliary MCP modules to simultaneously denoise video chunks at multiple future temporal horizons (next$^1$, next$^2$, next$^3$ chunks). These MCP modules form a causal chain across prediction depths, where intermediate features fused from multiple layers of the main model are leveraged to predict future dynamics, allowing near-future predictions to inform farther-future ones and providing dense multi-scale temporal supervision back to the main model. During training, the MCP modules significantly accelerate convergence and improve converged accuracy, especially at high frame rates: at 50 fps, Next Forcing achieves a 93.1% relative improvement over LingBot-VA at 5k training steps and 2.3x faster convergence, and establishes new state-of-the-art results on the RoboTwin benchmark (94.1/93.5% on Clean/Random). At inference, the MCP modules can be retained to predict the next video chunk in parallel with the current one, achieving 2x inference acceleration. Next Forcing also demonstrates significant improvements on PhyWorld, a benchmark evaluating adherence to physical laws in video generation, and over 50% FVD reduction on general video pretraining.

Problem

• autoregressive video generation
• training convergence
• inference speed
• high frame rates
• future dynamics

Innovation

• Multi-Chunk Prediction
• Causal World Modeling
• Autoregressive Video Generation
• Temporal Supervision
• Parallel Inference