In this work, I experimented with using an image editing model (Z-Image) that only predicts the end frame of an action chunk. On LIBERO, it already achieves comparable results. I believe it could…