Original title: One Editor, Many Edits: A Unified Training-Free Framework for Diverse Video Editing
Adheesh Sunil Juvekar, Onkar Kishor Susladkar, Kiet A. Nguyen, Muntasir Wahed +5 · September 2026
Editing videos using words or reference pictures is tricky. Most computer tools only handle one specific type of edit, making it hard to create a single system that does both well. Researchers built EditVid to solve this. It works without extra training by using three clever tricks. First, it tracks nearby video frames to keep motions smooth. Second, it matches features across distant frames to keep an object looking identical throughout the clip. Third, it gently blends edits so changes only touch the target area. EditVid handles tasks like changing styles, swapping subjects, and adding objects. In tests, it clearly beat other similar training-free tools and won a user study against 7 competitors with a 51.8% overall preference.
Imagine you're editing a family video and want to change the background in some scenes while keeping everyone's faces the same. Normally, you'd edit each clip separately, which could lead to mismatches and take a lot of time. Here, the system automatically handles diverse edits in a unified way, ensuring that changes are localized and identities are preserved throughout the video.
For video creators, this means they can perform a wide range of edits using just one tool, without needing to train it. This simplifies the editing process and makes professional-looking results more accessible.
The abstract does not explicitly state limitations, but it focuses on a training-free approach and evaluates on specific benchmarks, which may not cover all possible video editing scenarios.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
One Editor, Many Edits: A Unified Training-Free Framework for Diverse Video Editing — Adheesh Sunil Juvekar, Onkar Kishor Susladkar, Kiet A. Nguyen et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.