⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

Wonder: Video World Model Done Better

arXiv 2026 41.4 method, system

TLDR

Wonder is a real-time, camera-controllable video world model with novel camera conditioning, memory mechanism, and training strategy for interactive long-term world exploration.

Reasoning

The paper presents a strong system-level co-design for interactive video world modeling, but the abstract lacks explicit mention of quantitative evaluations or real-world benchmarks, making its empirical validation unclear. The method appears innovative in camera control and memory retrieval, yet the absence of comparison to existing baselines limits assessment of its claimed improvements.

Read-first score

Read-first score 41.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 55.

Recency 6%
100

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

Topical relevance 29%
78.6

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 18%
40

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=none

Reproducibility 18%
30

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

Citation impact 18%
0

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

Citation velocity 12%
0

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 398.

Keyword Scores

world model
10
video world model
10
generative world model
9
interactive world model
9
world simulator
8
world dynamics prediction
8
model-based reinforcement learning world model
1

Deep Analysis

Innovations

  • Camera conditioning via dense coordinate field renderings that provide spatially aligned motion and orientation cues as visual evidence.
  • Efficient sparse attention-based memory mechanism for fast and precise retrieval over a growing generation context, independent of context length.
  • Techniques to rectify the self-forcing-style distillation pipeline, improving the student model's ability to respect control signals, maintain diverse generation modes, and preserve long-term memory from the teacher.

Methodology

Wonder is a video world model that co-designs camera control, memory, and training. It uses a dense coordinate field rendering for camera conditioning, a sparse attention memory mechanism for efficient context retrieval, and a distillation training strategy from a teacher model with rectifications to improve control, diversity, and memory.

Key Results

Wonder synthesizes diverse, minute-scale videos at 16 FPS while preserving coherent geometry, appearance, and dynamics across long rollouts, and supports video-conditioned re-shooting of dynamic scenes in real time.

Resources

Paper

Tags