Video World Models · 2026-11-04
Compression Forcing: Hierarchical History Tokenization Enables Flexible World Models
Siqiao Huang, Partha Kaushik, Michael Chen, Hengkai Pan, Kaiwen Geng, Omar Chehab, Fernando Moreno-Pino, Max Simchowitz
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors propose Compression Forcing (CF), a diffusion-based world-model framework that aligns history compression with predictive utility. It uses Dynamics-Aware Tokenization (DAT) to encode history as a hierarchical token sequence ordered by dynamics saliency, and Hierarchical History Guidance (HG-H) to amplify dynamics-salient history at inference via score composition over variable-length prefixes.
Why it matters
CF aims to avoid tokenization/conditioning borrowed from video generation, which emphasizes perceptual realism rather than predictive salience needed for world-model planning. By learning saliency-ordered history tokens and enabling inference-time selective amplification, the approach targets improved future prediction, physical realism, and downstream decision-making.
Method
- Dynamics-Aware Tokenization (DAT): learn a hierarchical token sequence for history ordered by dynamic saliency using a schedule-weighted combination of static reconstruction and dynamic prediction objectives, with causal slot-token attention.
- Diffusion-based world model backbone: condition next-frame prediction on actions and arbitrary-length prefixes of DAT token sequences (via sliding-window DAT encoding and prefix truncation masks).
- Hierarchical History Guidance (HG-H): inference-time score composition using full-history and truncated-prefix conditional scores to amplify dynamics-relevant history while suppressing visual nuisances. (guided by budget vectors).
Abstract (from arXiv)
Effective world modeling requires compressing salient history information that facilitates future prediction. However, most existing approaches inherit tokenization and conditioning mechanisms from video generation, where objectives are centered around perceptual realism rather than predictive utility. This discrepancy limits the world model's effectiveness, especially in environments dominated by visual nuisances or task-irrelevant background details. To this end, we introduce Compression Forcing (CF), a general framework that aligns history compression with the predictive objective of world modeling. Compression Forcing features Dynamics-Aware Tokenization (DAT), a novel tokenization method that compresses history observations into a sequence ordered by decreasing dynamics saliency. Built on a diffusion backbone, this structure enables Hierarchical History Guidance (HG-H), allowing selective amplification of dynamics-salient information during inference. Extensive experiments across various environments demonstrate improved prediction fidelity, physical realism, and downstream decision-making performance, establishing Compression Forcing as a highly flexible and effective approach for world modeling.
Related papers
- History Forcing: Distilling History-Dependent Continuations for Long-Horizon Video World Models
- GROME: GROunding Factorized MEmory into Video World Models
- Reading, Writing, Sharing, and Interacting with the World through Video
- Rethinking Video Autoencoders for Interactive World Models
- World Models with Configurable Perception
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First announced by arXiv: 2026-11-04 · Indexed by Pub-AI: 2026-11-04 UTC
Compression Forcing: Hierarchical History Tokenization Enables Flexible World Models
First announced by arXiv: 2026-11-04 · Original record ↗
- arxiv
- 2611.04028
Explicit source versions (1)
- v1 · Date not supplied
- Preprint available
Source date: 2026-11-04 · First detected: 2026-11-04 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.04028v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv
Effective world modeling requires compressing salient history information that facilitates future prediction. However, most existing approaches inherit tokenization and conditioning mechanisms from video generation, where objectives are centered around perceptual realism rather than predictive utility. This discrepancy limits the world model's effectiveness, especially in environments dominated by visual nuisances or task-irrelevant background details. To this end, we introduce Compression Forcing (CF), a general framework that aligns history compression with the predictive objective of world modeling. Compression Forcing features Dynamics-Aware Tokenization (DAT), a novel tokenization method that compresses history observations into a sequence ordered by decreasing dynamics saliency. Built on a diffusion backbone, this structure enables Hierarchical History Guidance (HG-H), allowing selective amplification of dynamics-salient information during inference. Extensive experiments across various environments demonstrate improved prediction fidelity, physical realism, and downstream decision-making performance, establishing Compression Forcing as a highly flexible and effective approach for world modeling.
