HelloWorld: Enabling Socially Interactive Characters in Video World Models

arXiv 2026 67.1 method, system

TLDR

HelloWorld enables socially interactive characters in video world models via self-distillation and training-free cross-attention modulation.

Reasoning

The paper presents a novel method for adding social interaction to video world models, with a self-distillation pipeline and a benchmark. Strengths include clear contribution and evaluation; weaknesses are limited interaction types and lack of full world dynamics.

Read-first score

Read-first score 67.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 8%

100

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

Methodology quality 25%

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

Topical relevance 42%

64.3

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=github

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 150.

Keyword Scores

interactive world model

video world model

world model

generative world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Video world model enabling social interaction with characters via button press
Self-distillation pipeline that finetunes on self-synthesized data containing both social interactions and camera motion, learning camera-pose conditioning without degrading interaction quality
Training-free inference module that modulates cross-attention masks of DiT to temporally localize character responses to the press window
HelloWorldBench: a 400-sample benchmark with three social interaction metrics and three conventional metrics

Methodology

HelloWorld is a video world model that allows users to prompt characters to respond with a button press. It uses a self-distillation pipeline where the model generates training clips with social interactions and camera motion, then finetunes on them to learn camera-pose conditioning. At inference, a training-free module modulates the DiT's cross-attention masks so that the interaction prompt attends only to frames within the button press window, localizing the response temporally.

Key Results

HelloWorld surpasses various baselines in interaction quality on HelloWorldBench while maintaining state-of-the-art picture aesthetics and camera-pose following.

Resources

Paper