Answer
Caption instructions can also come from a selected Text hashtag. For a request covering several existing videos, select the intended loaded video messages before sending. A request accepts up to four uploaded videos, ten minutes each, twenty minutes total and 256 MB combined, subject to upload limits and processing time.
GIFs, Clips, arbitrary external video links, multiple audio streams and burned-in subtitles are unsupported. Silent videos produce no speech captions. Timing and translation can contain mistakes. You receive downloadable subtitle files and a short speech summary.
Generation uses measured Text tokens for summary and translation, with transcription included and no separate Video or Audio charge. Failed publication is not charged. Playing or downloading ready captions adds no AI request. Asking how subtitles work does not create them.