Where the transcript comes from
X generates captions for most videos with speech. They're the subtitles you see when a video autoplays on mute. This tool reads that caption track directly, so you get the same words X shows, with X's own timing.
Some videos have no captions: older uploads, very new posts X hasn't processed yet, and some languages. For those, the tool listens to the audio and writes the transcript with AI speech-to-text. It works for videos up to 15 minutes and still gives you timestamps. The result card always tells you which source was used.
