ImageWAM: Do World Action Models Really Need Video Generation, or Just Image Editing?
Quick Answer
ImageWAM presents a novel approach to World Action Models (WAMs) by utilizing image editing instead of video generation, achieving 1/6 the FLOPs and 1/4 the latency of traditional video-based models.
Quick Take
This method enhances action prediction accuracy by focusing on relevant visual changes, outperforming standard VLA baselines and competitive WAMs without additional policy pretraining across various experiments.
Key Points
- ImageWAM leverages pretrained image editing models for robot action prediction.
- It reduces computational costs to 1/6 FLOPs and latency to 1/4 of video-based WAMs.
- The model focuses on action-relevant visual differences rather than irrelevant details.
- ImageWAM outperforms standard VLA baselines in various simulator and real-world tests.
- Attention analysis shows editing caches target task-relevant change regions.
Paper Resources
📖 Reader Mode
~2 min readAbstract:World Action Models (WAMs) commonly rely on video generation to bridge visual world modeling and robot control. However, video-based WAMs face three coupled limitations: dense multi-frame future tokens make inference costly, full video prediction spends capacity on action-irrelevant temporal and appearance details, and long-horizon future imagination may introduce errors that mislead action prediction. These issues raise a simple question: Does world action model really need video generation? We propose ImageWAM, a simple WAM framework that repurposes pretrained image editing models for robot action prediction. In contrast to video generation, image editing provides a better-matched prior: it only needs to model a target-frame transformation, focuses on action-relevant current-to-target visual differences, and grounds task instructions to localized visual changes through edit pretraining. In practice, ImageWAM does not decode the target frame at inference time; instead, it conditions a flow-matching action expert on the KV caches produced by image-editing denoising, using them as a compact world-action context. ImageWAM outperforms standard VLA baselines and matching competitive WAMs without additional policy pretraining across different simulator and real-world experiments. It also reduces FLOPs to 1/6 and latency to 1/4 of video-based WAMs. Attention analysis further shows that editing caches focus on task-relevant change regions, supporting image editing as an effective alternative to video-based world-action modeling.
| Comments: | Project Page: this https URL |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO) |
| Cite as: | arXiv:2606.19531 [cs.CV] |
| (or arXiv:2606.19531v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2606.19531 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yuyang Zhang [view email]
[v1]
Wed, 17 Jun 2026 19:25:28 UTC (3,920 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.