Mise-en-Video adds 3D staging control for multi-shot video
Original title: Mise-en-Video: Fine-Grained 3D Control for Multi-Shot Video Generation
Training uses a 160,000-video set with hierarchical multi-shot text and oriented 3D bounding boxes.
The paper proposes Mise-en-Video, a multi-shot text-to-video framework that lets creators stage shots in 3D space and then generate the video. It combines hierarchical attention for narrative consistency with 3D-aware layout guidance for per-shot spatial control. The authors built annotation pipelines and a 160,000-video dataset, and report that MeV beats strong baselines on quality, inter-shot consistency, and subject-level controllability.
- Existing multi-shot methods are described as consistent but weak on control.
- Guidance uses oriented 3D bounding boxes for each shot.
- The dataset pairs hierarchical multi-shot texts with those boxes.
- Evaluation covers video quality, consistency, and subject-level control.
- The short name used in the paper is MeV.
From the original article
1 min read at arxiv.orgRecent advances in text-to-video generation have enabled high-fidelity single-shot synthesis, yet practical storytelling creation requires multi-shot videos with both strong inter-shot consistency and fine-grained controllability. Existing multi-shot video generation approaches achieve strong consistency but still fall short in controllability.
By Yingshuang Zou · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.