Volcano Engine LAS introduces Video Editing Enhanced 2.0
Volcano Engine LAS has introduced the Video Editing Enhanced 2.0 operator, a Seedance-backed workflow for restyling complete videos or replacing selected props, products, backgrounds and people. According to the report, users provide a source video, editing instructions and reference images, while the workflow is designed to preserve the original action, camera movement, composition and timeline. The report presents short-form drama and e-commerce marketing as the main applications, with all showcased source materials generated by AI. It also outlines the asynchronous API process, pricing and limitations. Stability may decline for very short shots, frequent cuts, multiple similar-looking people, close-ups, heavy occlusion and large differences between old and new object shapes; subtitles and text may also contain garbled characters. The report recommends retries and manual refinement for such cases.
The operator packages video understanding, key-frame processing, structural-information extraction and final generation into one asynchronous API, allowing existing footage to be restyled or adapted without reshooting. The report highlights practical use cases including converting short dramas into animated formats and updating products or backgrounds in e-commerce videos.
如果说生成一段新视频,考验的是模型的创造力;那么在已有视频上做局部编辑,考验的则是另一种能力 —— 对“不该动的部分”的控制力。
一个典型的例子是 漫剧****改编:真人短剧希望整体转绘为动漫风格,逐帧人工重绘的成本难以承受;直接调用生成模型,又常出现画风中途回退、人物动作走样的问题。
类似的难题也摆在 电商团队 面前。大促期间主推款上新,存量视频的动作、运镜与节奏均已成熟,唯独展示的商品需要更新;而模型重做的结果,往往是商品与原有镜头语言一起被改变,修改一轮后成本随之上升。
两类场景的共同诉求高度一致:编辑范围严格限定在局部,画面其余部分保持不动。
火山引擎 AI 数据湖服务(LAS)的 视频编辑增强版 2.0 算子,正是为满足这一诉求而设计。依托于 Seedance 视频生成能力,提供面向业务场景的编辑模板:只需输入原视频、编辑指令与参考图,便可在完整保留原片动作、镜头、构图与时间线的基础上,完成全片风格转绘,以及道具、商品、背景与人物的定向替换。影视短剧与电商营销,正是其当前最主流的两大应用场景。