Generating Long Videos of Dynamic Scenes

arXiv 2022 32.6 method

TLDR

A video generation model that redesigns temporal latent representation and uses two-phase training to generate long videos with consistent object motion, camera viewpoint, and new content.

Reasoning

The paper addresses a key limitation in video generation—long-term temporal consistency—by introducing a redesigned temporal latent representation and a two-phase training strategy, and it contributes new benchmark datasets. However, the abstract does not report quantitative results or comparisons, and the proposed benchmarks may not fully capture real-world complexity.

Read-first score

Read-first score 32.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 9.

Recency 8%

56.5

Uses a gentle age decay so recent papers surface without erasing older foundations. 2022

Reproducibility 25%

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

Methodology quality 25%

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

Topical relevance 42%

12.9

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

Field roles

Candidate

Rank sensitivity

Stability: volatile; rank range: 52.

Keyword Scores

world dynamics prediction

video world model

world model

generative world model

world simulator

interactive world model

model-based reinforcement learning world model

Resources

Paper