arXiv:2606. 07687v1 Announce Type: cross Abstract: Video world models are increasingly used to provide predictive visual representations, yet it remains unclear which pretraining signals induce action-relevant structure in their latent spaces.
The paper introduces FRED, a framework that self‑distills future video representations for World Action Models (WAMs) by predicting these representations from noise using ground‑truth future observations as targets. By conditioning on historical observations, FRED improves the quality of future representations obtained in a single denoising step, thereby enhancing action prediction efficiency. Experiments on real‑world manipulation tasks show that FRED outperforms existing WAMs, especially on long‑horizon, out‑of‑domain, and zero‑shot scenarios, achieving a 50.0% success rate on tasks only seen in human videos versus 19.4% for the best competitor.
StateCraft introduces a new architecture that separates a persistent, updatable abstract state from the visual state in video world models. By pairing a language model with a video diffusion model, it tracks entity changes over time and conditions visual generation on this abstract state. The approach is evaluated on generative interfaces and a new benchmark, showing improved generation quality, memory usage, and controllability, while enabling direct editing of unseen states and extending to spatial 3D environments.
arXiv:2606. 02522v1 Announce Type: cross Abstract: Video multimodal large language models (MLLMs) have made rapid progress on general and long-form video understanding, yet their ability to preserve brief answer-critical visual evidence remains underexplored.
arXiv:2606. 10620v1 Announce Type: cross Abstract: Image generation models now produce high-quality static images, yet their ability to represent how a visual world changes over time remains poorly understood.
arXiv:2605.29858v3 Announce Type: replace Abstract: Temporal action localization (TAL) requires recognizing the target event and localizing its start and end times precisely in untrimmed videos. Rece...
arXiv:2602.18742v2 Announce Type: replace-cross Abstract: Synthetic data generated by video generative models has shown promise for robot learning as a scalable pipeline, but it often suffers from no...
arXiv:2608. 11605v1 Announce Type: new Abstract: World Action Models (WAMs) couple future visual prediction with robot action generation, enabling policies to model how the physical world evolves during interaction.
arXiv:2609.38377v1 Announce Type: new Abstract: Evaluating the physical consistency of generated videos remains a fundamental challenge. Existing approaches rely on off-the-shelf vision-language mode...
STRAND is a new benchmark that tests multimodal large language models’ ability to track objects, their states, and relationships over time in videos. It evaluates intermediate reasoning by breaking queries into sub‑questions and uses Faithful Accuracy to ensure all parts of an answer are correct. The authors also propose an object‑centric framework that builds structured trajectories and shows reduced hallucinations and better temporal consistency compared to existing models.
arXiv:2610.00812v1 Announce Type: cross Abstract: Video generation has rapidly progressed from short, low-quality clips to high-resolution, long-duration sequences with complex spatiotemporal dynamic...
arXiv:2610.35148v1 Announce Type: new Abstract: Recent advances in generative video models have improved visual realism, making detection increasingly dependent on subtle temporal inconsistencies. Ho...