Generative World Renderer at the Speed of Play

arXiv 2026 37 method, system

TLDR

A real-time generative world renderer that synthesizes RGB frames from physics engine states, achieving 31.54 FPS via autoregressive streaming and distilled codecs.

Reasoning

Strengths: Achieves real-time performance (31.54 FPS) for generative world rendering, preserving scene structure and enabling interactive play. Weaknesses: Limited novelty as a technical report on optimization; evaluation relies on G-buffer streams and does not address broader world model or RL applications.

Read-first score

Read-first score 37, weighted from topical fit, citation, graph, method, reproducibility, and recency signals.

Recency 6%

100

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

Topical relevance 29%

52.9

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

Methodology quality 18%

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

Reproducibility 18%

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

Citation impact 18%

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

Citation velocity 12%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 185.

Keyword Scores

interactive world model

generative world model

video world model

world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Reformulation of the generative world renderer as a few-step autoregressive streaming model to achieve real-time rendering speed
Introduction of lightweight distilled codecs for efficient latent encoding and frame reconstruction
Preservation of G-buffer and text-prompt interfaces while enabling continuous rendering over unbounded input streams

Methodology

AlayaRenderer-Flash is a distilled student model of the original AlayaRenderer, reformulated as a few-step autoregressive streaming process with lightweight distilled codecs. It retains the teacher's G-buffer and text-prompt interfaces and is evaluated on G-buffer streams across content preservation, temporal consistency, cross-window stability, prompt controllability, and runtime efficiency.

Key Results

AlayaRenderer-Flash accelerates rendering from 0.56 FPS to 31.54 FPS, and when integrated with a physics engine, delivers a fully playable generative world at 30 FPS while preserving the core rendering capabilities of the teacher model.

Resources

Paper