Honeycomb: Constant-Size Scene Memory Representation for Video World Models

arXiv 2026 34.5 method

TLDR

Introduces Honeycomb, a video world model with HexMemory, a fixed-size low-rank scene memory, enabling constant-storage long-horizon generation with strong quality and consistency.

Reasoning

The paper presents a novel fixed-size memory representation for video world models, addressing storage growth and demonstrating results on benchmarks. However, the abstract provides limited discussion of limitations or interactive/RL applications, and the core claims rely on benchmark results not visible in detail.

Read-first score

Read-first score 34.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 40.

Recency 5%

100

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

Topical relevance 25%

57.1

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 15%

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

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,github

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 306.

Keyword Scores

video world model

world model

generative world model

world dynamics prediction

world simulator

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

HexMemory: a low-rank fixed-size scene memory using six spatial and spatiotemporal planes
Feed-forward writer that maps each generated chunk into new plane features, warps previous planes while preserving dimensions, and fuses via confidence-weighted pooling and learned residual correction
Avoids per-scene optimization and repeated processing of full history by processing only observations from the new chunk
Reader that retrieves latents from HexMemory to condition subsequent video generation

Methodology

Honeycomb is a video world model built on HexMemory, a low-rank fixed-size memory representation with six spatial and spatiotemporal planes. A feed-forward writer maps each generated chunk into new plane features, warps previous planes while preserving their dimensions, and fuses them with new features using confidence-weighted pooling and learned residual correction. A reader retrieves latents from HexMemory to condition subsequent generation, and the model is evaluated on WorldScore and RealEstate10K.

Key Results

Experiments on WorldScore and RealEstate10K demonstrate strong video generation quality and robust revisit consistency while HexMemory feature storage remains constant throughout generation.

Resources

Paper