EditVid Framework for Video Editing
One Editor, Many Edits: A Unified Training-Free Framework for Diverse Video Editing
EditVid outperforms existing training-free video editing methods on FiVE benchmark and is preferred by users, offering a unified solution for diverse editing tasks.
EditVid is a training-free framework that combines sparse causal memory, correspondence-based post-attention token injection, and soft latent blending.
It supports instruction-guided and reference-guided edits including style transfer, attribute modification, object insertion, part-level editing, and subject replacement.
On FiVE benchmark, EditVid achieves 78.16 FiVE-Acc, compared with 58.
95 for the strongest training-free baseline.A user study shows 51.
8% preference for EditVid over 7 competing methods.
One Editor, Many Edits: A Unified Training-Free Framework for Diverse Video Editing
Video editing spans diverse editing paradigms, yet achieving high-quality instruction-guided and subject-guided editing within a single unified framework remains challenging. We introduce EditVid, a training-free framework combining sparse causal memory for local coherence, correspondence-based post-attention token injection for long-range identity preservation, and soft latent blending for edit locality. The same framework supports instruction-guided and reference-guided edits, including style transfer, attribute modification, object insertion, part-level editing, and subject replacement. On FiVE, EditVid achieves 78.16 FiVE-Acc, compared with 58.95 for the strongest evaluated training-free baseline, while obtaining competitive results on IVEBench. A user study further shows a 51.8\% overall preference for EditVid over 7 competing methods.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include EditVid, video editing, FiVE, IVEBench.
Source attribution: curated from arXiv cs.AI (@Adheesh Sunil Juvekar, Onkar Kishor Susladkar, Kiet A. Nguyen, Muntasir Wahed, Nabeel Bashir, Xiaona Zhou, Tianjiao Yu, Vedant Shah, Ismini Lourentzou).
