Hunyuan-GameCraft-2: Instruction-following Interactive Game World Model

arXiv 25.11 2025 69.1 method, system, application

TLDR

Introduces an instruction-following interactive game world model using natural language, keyboard, or mouse control, with a benchmark for evaluation.

Reasoning

The paper presents a novel paradigm for interactive game world modeling with flexible instruction-driven control, supported by an automated dataset creation process and a dedicated benchmark. Strengths include the integration of diverse input modalities and causal grounding, while weaknesses may involve scalability and reliance on a large MoE model.

Read-first score

Read-first score 69.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 51.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

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

Topical relevance 42%

72.9

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

Reproducibility 25%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 39.

Keyword Scores

interactive world model

world model

generative world model

video world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Instruction-driven interaction paradigm for generative game world modeling, replacing fixed keyboard inputs with natural language, keyboard, or mouse signals
Automated process to transform large-scale unstructured text-video pairs into causally aligned interactive datasets
Text-driven interaction injection mechanism for fine-grained control over camera motion, character behavior, and environment dynamics
Interaction-focused benchmark InterBench for comprehensive evaluation of interaction performance

Methodology

The model is built upon a 14B image-to-video Mixture-of-Experts (MoE) foundation model, incorporating a text-driven interaction injection mechanism for fine-grained control. An automated process transforms large-scale unstructured text-video pairs into causally aligned interactive datasets. Evaluation is performed using a newly introduced interaction-focused benchmark, InterBench.

Key Results

The model generates temporally coherent and causally grounded interactive game videos that faithfully respond to diverse free-form user instructions such as 'open the door', 'draw a torch', or 'trigger an explosion'.

Resources

Paper Project