光郡

You can transcribe talking-head videos into timestamped transcripts, identify semantic beats needing B-roll, match T01–T21 packaging templates, choose simple/rich/balanced prompt rules, generate MiniMax-H3 shots after approval, and auto-edit them onto the source video by timestamp.

It targets video directors, content editors, and口播 creators—aligned with Guangjun's TVC director background—who need to plan B-roll, apply packaging templates without background music (sound effects only), and produce finished edited videos from talking-head footage.

The skill accepts a transcript or talking-head video, optionally transcribes it with timestamps, selects v1/v2/v3 prompt rules, identifies B-roll beats, matches packaging templates, requests truthful source assets, generates MiniMax-H3 shots only after explicit approval, then edits them onto the video by timestamp.

You need a talking-head video or an existing transcript as input. The skill handles transcription optionally, but MiniMax-H3 shot generation requires your explicit approval before any shots are created and edited back onto the source footage.