How the picker decides
After the recording is transcribed, the full timestamped transcript is handed to a language model with a specific brief: behave like an editor choosing stand-alone clips for a vertical feed. It is told to favour a clear hook in the opening sentence, one idea developed to a conclusion, and a natural end. It is told to prefer stories, strong opinions, surprising facts, practical steps and emotional turns, and never to begin a window in the middle of a sentence or end it before the thought resolves.
The model returns candidate windows with a start time, an end time, a short title and a one-line reason. The clipper then snaps each window to the nearest word boundaries, stretches anything under twenty seconds forward to the next sentence end, trims anything over the ceiling at a word boundary, and removes overlaps. What remains is rendered.
Titles you can use as-is
Each highlight gets a title of at most eight words, written as a hook rather than a summary. It appears on the clip card in your library and on the public page if you choose to share a clip. Many people paste it straight into the post caption; it is written to work that way.
What makes a good source for highlights
The picker works from words, so the best sources are conversational: interviews where a guest tells a story, solo episodes built around opinions, lectures with worked examples, keynotes with a clear argument. Recordings with long stretches of silence, music, or screen activity without narration give the picker little to work with, and it will say so rather than invent something.
If you want more variety from one source, ask for five clips with a lower length ceiling. The selection will spread across the recording instead of clustering around the one strongest section.
From highlight to finished short
A highlight on its own is just a pair of timestamps. The clipper turns it into something postable: cut precisely, reframed to 9:16 with your chosen crop mode, captions rendered from the same word timings the picker used, and a progress bar along the bottom. Each short is written to your library with its title, its position in the source and the excerpt of transcript it covers, so you can see exactly why it was chosen.
Privacy and cost
Transcripts are checked against a content filter before any rendering happens; sexual content is refused and the credits are returned. The source upload is deleted automatically after twenty-four hours. A full highlight job is forty credits, however many clips it produces, and a failed job that is not your fault is refunded automatically.