MoviEdit: Long-form Multi-Shot Video Editing Made Easy
Suhyeon Lee, Zhengang Li, Soo Ye Kim, Peihao Zhu, Jong Chul Ye, Chun-Hao Paul Huang, Duygu Ceylan, Yan Kang and 1 others
Adobe · KAIST
Published November 2, 2026 · on the feed November 4, 2026 · ▲ 74 · arXiv 2611.02737
The abstract
Long-form multi-shot video editing requires an intended edit to remain coherent across abrupt changes in scene, viewpoint, and subject appearance, while processing substantially longer visual context than conventional single-shot editors. We introduce MoviEdit, an end-to-end framework for scalable long-form multi-shot video editing that jointly addresses the challenges of data, modeling, and evaluation. To provide cross-shot supervision data at scale, we develop automatic data-generation pipelines spanning synchronized multi-view footage, real-world multi-shot videos, and generated multi-shot content, yielding 87K paired editing videos without manual annotation. Architecturally, MoviEdit employs a cascaded bidirectional-autoregressive design: a lightweight bidirectional stage first establishes a globally coherent editing draft over a spatiotemporally compressed representation, which is then autoregressively propagated and refined over the full video. We further introduce MoviEdit-Bench, a dedicated benchmark for systematic evaluation of long-form multi-shot video editing. Extensive experiments show that MoviEdit substantially outperforms existing open-source models and performs on par with industry frontier models across text-guided and image-guided editing. Visit our project page for complete results and to access the benchmark: https://multi-shot-edit.github.io/
The Brief
2:01A short podcast summary of the paper. Written and voiced by AI.
Transcript
MoviEdit. Long-form Multi-Shot Video Editing Made Easy. By Suhyeon Lee and eight others, from Adobe and KAIST. Published on the second of November.
Editing a long video made of many shots is harder than editing a single clip. The intended edit has to stay coherent across abrupt changes: a cut to a new scene, a new viewpoint, a subject whose appearance shifts. At the same time, the editor has to process far more visual context than a conventional single-shot editor ever sees. Data, modeling and evaluation all become problems at once.
MoviEdit is an end-to-end framework that takes on all three together. Its aim is long-form multi-shot editing that scales, and its answer comes in three parts: a way to make supervision data, an architecture, and a benchmark.
For data, the authors build automatic generation pipelines that draw on three sources: synchronized multi-view footage, real-world multi-shot videos, and generated multi-shot content. Together they yield eighty-seven thousand paired editing videos, with no manual annotation, giving supervision that spans across shots.
The architecture is a cascaded bidirectional-autoregressive design. A lightweight bidirectional stage works first, over a spatiotemporally compressed representation of the video, and establishes a globally coherent editing draft. That draft is then propagated autoregressively and refined over the full video. Global coherence is settled cheaply on the compressed form, and detail is filled in afterwards at full length. For evaluation, the authors introduce MoviEdit-Bench, a dedicated benchmark for systematic assessment of long-form multi-shot editing.
Extensive experiments cover both text-guided and image-guided editing. According to the abstract, MoviEdit substantially outperforms existing open-source models, and performs on par with industry frontier models.
What makes this a framework rather than a single model is that each of the three bottlenecks gets its own answer. Without paired data spanning shots there is nothing to learn from, without a coherent global draft the edit drifts between cuts, and without a dedicated benchmark there is no way to tell. MoviEdit addresses them together.
Paper Quest · 7 chapters · about 41 min
Play this paper
Learn it by doing: make the calls the authors made, guess the result before you see it, and defend it in a viva graded on your own words. Chapter 1 is right here.
Chapter 1 🧩 · The problem
One edit, many shots, and every cut tries to break it.
You are asked to edit a long video cut from several shots: remove one man in a tactical suit everywhere he appears. The camera angle, the scene and his look change at every cut.
The road ahead
- 1 · The problem ~7 min Editing one subject across a long, many-shot video breaks today's editors: they lose the subject at cuts, choke on length, and lack training pairs.
- 2 · What came before ~3 min
- 3 · The key idea ~9 min Make cross-shot training pairs automatically, draft the edit cheaply on a compressed video, refine it at full length, and test on a dedicated benchmark.
- 4 · Choose your own research ~3 min
- 5 · Build the method ~8 min Build paired data, draft the edit bidirectionally on a subsampled video, render it causally chunk by chunk, then score it on MoviEdit-Bench.
- 6 · The experiment ~6 min You’ll guess: Across text-guided and image-guided editing, how did MoviEdit compare with industry frontier models?
- 7 · The viva ~5 min Three questions from the examiners, answered in your own words. Win it for the paper’s trophy.
Free, with Google or an email address. Your stars and progress are kept.