Aurora: Unified Video Editing with a Tool-Using Agent

Researchers introduced Aurora, an agentic video editing framework that pairs a tool-augmented vision-language model with a unified video diffusion transformer.

Aurora can interpret underspecified editing requests, fill in missing details, and generate structured edit plans before video synthesis.

The system improves editing quality on AgentEdit-Bench and other benchmarks, especially for ambiguous requests.

It addresses a common real-world gap in video editing: users often leave out information that models need to produce good edits.