Paper Overview
Field: Computer Vision Authors: Zhenghong Zhou, Zhe Lin, Jiebo Luo, Yuqian Zhou Published: 2026-10-06 arXiv: 2610.08779
Summary
Current video editors can insert objects but often struggle to make them participate in interactions such as being picked up or manipulated. The authors introduce ALIVE, a framework that makes inserted objects "alive" through coherent interactions with the source video's contents, using an edited first frame and an instruction naming only the added object.
Key Contributions
- 35,800 curated editing pairs, combining 3D-rendered, model-generated, and real-world videos with general editing pairs from ROSE. Each pair differs in the target object's presence while preserving the surrounding action, teaching the editor coordinated object behavior and source preservation.
- A vision-language model (VLM) trained to predict interaction guidance from the same inputs, removing the need for extra user input.
- The ALIVE-interaction benchmark, assessing interaction fidelity, source preservation, and visual coherence under a unified VLM protocol.
- Without VLM guidance, ALIVE improves overall scores by 43.9% and 4.4% over the strongest evaluated baselines on two benchmarks.
- VLM-predicted guidance further improves the ALIVE-interaction score by 0.95 points without additional user input.
Results
Original Abstract
> Current video editors can insert objects but often struggle to make them participate in interactions such as being picked up or manipulated. We introduce ALIVE, a framework that makes inserted objects "alive" through coherent interactions with the source video's contents, using an edited first frame and an instruction naming only the added object. We curate 35,800 editing pairs combining 3D-rendered, model-generated, and real-world videos with general editing pairs from ROSE. Each pair differs in the target object's presence while preserving the surrounding action, teaching editors coordinated object behavior and source preservation. We further train a vision-language model (VLM) to predict interaction guidance from the same inputs. We introduce the ALIVE-interaction benchmark to assess interaction fidelity, source preservation, and visual coherence under a unified VLM protocol.
---
*Auto-collected on 2026-10-08*