ImageWAM: Do World Action Models Really Need Video Generation, or Just Image Editing?

ImageWAM replaces video generation in robot world models with single-frame image editing, predicting only the target frame rather than full temporal sequences. On robot manipulation benchmarks, this reduces inference cost while maintaining competitive action prediction accuracy by avoiding capacity waste on action-irrelevant temporal details. The approach repurposes pretrained image editing models without task-specific video generation pretraining.