It finds the parts of a long recording that stand on their own, about a minute each, and ticks the strongest three. You can tick others and move their edges.
It makes each one tall (9:16) with the speaker in frame, and writes the captions in. One clip at a time, so most phones have room for it. An older iPhone with a small screen may not, with a 1080p recording.
Each clip is an MP4 as tall as your recording: 608×1080 from a 1080p recording, 406×720 from 720p. It is not enlarged, so it is as sharp as what you recorded.
Each clip is up to 90 seconds. A longer moment ends on the last whole sentence before that.
Recordings up to four hours, with a picture. English. It does not guess what an audience will do with a clip.
It takes a while. On a recent computer it reads an hour of recording in about 7 minutes, and each clip takes about half as long as it plays. A phone is slower, and has not been timed yet.
Where your recording goes
Nowhere. It all runs in this page, on your device. There is no account and no upload.
The tool itself is downloaded the first time you use it, from where it is published. Your recording is not part of that.