Next Forcing: Causal World Modeling with Multi-Chunk Prediction

Hugging Face Daily Papers Papers
autoregressive-video-generation causal-world-modeling multi-chunk-prediction video-generation inference-acceleration

Summary

Next Forcing introduces a multi-chunk prediction framework for causal world modeling that accelerates training and inference for autoregressive video generation while improving accuracy and physical law adherence.

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.
Original Article
View Cached Full Text

Cached at: 06/10/26, 01:44 PM

Paper page - Next Forcing: Causal World Modeling with Multi-Chunk Prediction

Source: https://huggingface.co/papers/2606.11187

Abstract

Next Forcing introduces a multi-chunk prediction framework that accelerates training and inference for autoregressive video generation while improving accuracy and physical law adherence.

Autoregressive video generationhas emerged as a powerful paradigm forWorld 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 iterativevideo denoising. In this paper, we present Next Forcing, amulti-chunk prediction(MCP) framework forcausal world modelingthat 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 futuretemporal horizons(next^1, next^2, next^3 chunks). These MCP modules form acausal chainacross 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 densemulti-scale temporal supervisionback 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 2xinference acceleration. Next Forcing also demonstrates significant improvements onPhyWorld, a benchmark evaluating adherence to physical laws in video generation, and over 50%FVDreduction on general video pretraining.

View arXiv pageView PDFProject pageGitHub29Add to collection

Get this paper in your agent:

hf papers read 2606\.11187

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.11187 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.11187 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.11187 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Looking Ahead: Modeling Future Evolution in Autoregressive Video Generation

Hugging Face Daily Papers

The paper introduces Future Chunk Prediction (FCP), an auxiliary objective for autoregressive video diffusion models that predicts temporally separated chunks without intervening visual context, improving physical plausibility and reducing training compute. The approach requires no additional pretrained visual encoder and hints at more structured representation spaces useful for autoregressive world models.

Frechet Forcing: Training Long-Horizon Streaming World Models from Short Clips

Hugging Face Daily Papers

Fréchet Forcing 提出了一种无需教师模型和长视频的微调方法，利用自生成轨迹与真实数据统计量（生成锚定、上下文保真、现实匹配三项约束）训练流式视频世界模型，使其在超过 4000 帧的长时程推理中保持稳定并正确跟随动作指令。