Reproduce “ImVideoEdit: Image-learning Video Editing via 2D Spatial Difference Attention Blocks” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jiayang Xu, Fan Zhuo, Majun Zhang, Changhao Pan, Zehan Wang, Siyu Chen, Xiaoda Yang, Tao Jin and 1 more

ImVideoEdit edits videos from text instructions using paired images rather than paired video examples. Its Predict-Update spatial module estimates and refines changes, a text-conditioned gate controls the refinement, and pretrained 3D attention remains frozen. On a VLM-based benchmark, it scores 65.24 versus 56.79 for VACE-1.3B, but trails Kiwi-Edit-5B’s 71.13 and is not best overall.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.