Render, Don't Decode: Weight-Space World Models with Latent Structural Disentanglement

arXiv 2026 58.3 method

TLDR

NOVA uses weight-space implicit neural representations to render world states, eliminating decoder bottlenecks and enabling disentangled scene editing.

Reasoning

The paper introduces a novel framework that replaces traditional latent decoders with analytically rendered INRs, achieving compactness and zero-shot super-resolution. Its strengths include strong empirical validation on challenging datasets and unsupervised disentanglement of structural components. Weaknesses are the lack of explicit real-world deployment details and limited discussion of scalability beyond single GPU.

Read-first score

Read-first score 58.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.

Recency 6%

100

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

Topical relevance 29%

77.1

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

Citation impact 18%

72.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.72657854

Methodology quality 18%

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

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FrontierBridge

Rank sensitivity

Stability: volatile; rank range: 451.

Keyword Scores

world model

generative world model

video world model

world dynamics prediction

world simulator

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Representing world model state as weights and biases of an auxiliary coordinate-based implicit neural representation (INR) instead of opaque latent codes
Analytic rendering of the INR state to eliminate the decoder bottleneck, enabling compactness, portability, and zero-shot super-resolution
Disentanglement of structural scene components (background, foreground, inter-frame motion) without auxiliary losses or adversarial objectives
Distillation into a context-dependent video generator via an action-matching objective

Methodology

NOVA models the system state as the weights and biases of a coordinate-based implicit neural representation (INR), which is analytically rendered to produce observations without a heavy decoder. The framework uses an action-matching objective to distill the world model into a controllable video generator, and achieves structural disentanglement without auxiliary losses.

Key Results

The framework achieves strong controllable forecasting on several challenging datasets while operating on a single consumer GPU at approximately 40 million parameters, and enables zero-shot super-resolution and editable scene components.

Resources

Paper