ALIVE：面向首帧引导视频编辑的交互对齐物体插入

AI 摘要

研究者提出 ALIVE 框架，用于首帧引导的视频编辑，使插入的物体能与原视频内容产生连贯交互（如被拿起或操作）。他们构建了 35,800 对编辑数据，结合 3D 渲染、模型生成和真实视频，并训练 VLM 从相同输入预测交互引导。在 ALIVE-interaction 基准和通用视频物体插入基准上，ALIVE 的 Overall 指标分别比最强基线提升 43.9% 和 4.4%，VLM 引导进一步将交互分数提升 0.95 分。

正文节选

ALIVE: Interaction-Aligned Object Insertion for First-Frame-Guided Video Editing 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-

发布时间：—

抓取时间：2026-10-07 10:14

来源机构：Hugging Face