What is it about?

Given a user request (eg: "make a reel of my travel") and user assets (eg: travel videos shot on phone), our framework is an end-to-end system that is able to output the final requested short-video. It does so in a hierarchical manner: first it generates a high-level plan that defines the video arc, scene structure, pacing, and style, and then generates a low-level plan that maps scenes to specific clips, durations, effects, text, and narration. Lastly, it creates the desired video via our Plan2Video renderer which parses the low-level plan to edit the user assets into a final video.

Featured Image

Why is it important?

With social media videos becoming the dominant form of digital content, for content creators (such as ad campaigners and social media influencers), the ability to produce compelling short-form videos is now an essential skill. Producing such videos, however, is a demanding process. Beyond the technical barrier of mastering editing software, a creator must navigate a dense web of creative decisions: which clips to select, how to impactfully sequence them, what pacing to maintain, which transitions and effects to reinforce the story, and how to sustain viewer engagement from the first frame to the last. Creating an engaging social media video is therefore not a matter of simply stitching raw footage together, but a challenging, multi-level planning problem that is time-consuming and inaccessible to beginner and amateur creators alike. Currently, no approach allows a user to express creative intent in natural language and receive a structured, inspectable plan for assembling a final video from their own footage.

Read the Original

This page is a summary of: LoopEdit: Learning Data-driven Hierarchical Planners for Short-Video Assembly, November 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3767308.3834917.
You can read the full text:

Read

Contributors

The following have contributed to this page