LIVE: Long-horizon Interactive Video World Modeling

arXiv 26.2 2026 66.5 method

TLDR

LIVE introduces a cycle-consistency objective for long-horizon video world models, eliminating teacher-based distillation and achieving state-of-the-art performance.

Reasoning

The paper presents a novel cycle-consistency approach to bound error accumulation in autoregressive video world models, which is a clear strength. However, the abstract lacks explicit details on real-world datasets and does not discuss limitations or comparisons with baselines in depth.

Read-first score

Read-first score 66.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 60.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 42%

85.7

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=benchmark,experiment

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 331.

Keyword Scores

world model

interactive world model

video world model

generative world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Cycle-consistency objective that enforces bounded error accumulation without teacher-based distillation
Unified view encompassing different approaches to long-horizon video world modeling
Progressive training curriculum to stabilize training

Methodology

LIVE performs a forward rollout from ground-truth frames, then applies a reverse generation process to reconstruct the initial state. The diffusion loss is computed on the reconstructed terminal state, providing an explicit constraint on long-horizon error propagation.

Key Results

LIVE achieves state-of-the-art performance on long-horizon benchmarks, generating stable, high-quality videos far beyond training rollout lengths.

Resources

Paper