State World Models: Execute the Rules, Generate the Pixels
We present Stateful Harness, a framework that separates world evolution from video rendering to support stateful and rule-governed interaction.
ProofPaper ↗
Key points
- Existing world models can generate action-conditioned video from a single image.
- It maintains an evolving world state under specified rules, renders each resulting event with a frozen video model, and uses the generated clip to update the world state for the next turn.
- To evaluate rule compliance, we build WorldState-Bench, a benchmark of 50 counterfactual twin pairs that share an initial frame and action but differ in hidden state and required outcome.
- These results support separating rule execution from video rendering for world interactions with hidden state.
Sources (2)
- [1]State World Models: Execute the Rules, Generate the PixelsHugging Face Daily Papers · Oct 19, 12:00 AM
We present Stateful Harness, a framework that separates world evolution from video rendering to support stateful and rule-governed interaction.
Existing world models can generate action-conditioned video from a single image.
- [2]State World Models: Execute the Rules, Generate the PixelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 05:50 PM · same content
Extractive summary: sentences quoted from the sources.
Before this
- Oct 18, 2026Google Open-Sources Harness2, a Recursive Agent Harnessing Framework
- Oct 18, 2026World Model Interpretability: From Decoding Representations to Understanding Latent Worlds
- Oct 16, 2026Execution Before Memory: Diagnosing Memory Failures in Video World Models
- Oct 16, 2026OfficeTown: Generative Organizations for Evolving Agent Intelligence
- Oct 16, 2026Persistent Planar Memory for Video World Models
- Oct 16, 2026Language Models Learn to Improve Language Models