Getting a transcript of a video

A file you already have

Any MP4, MOV, MKV or WebM works, which covers a phone video, a screen recording, a lecture capture and an exported edit. Only the soundtrack is read. The picture is decoded for the preview and ignored by the model, so a 4K file and a 360p copy of it transcribe identically and the small one loads faster.

What comes back is prose rather than captions: paragraphs broken at the pauses in speech, each with the time it starts, as Markdown or plain text. Timestamps are in seconds from the start of the file, the same clock the video player uses, so a time in the transcript can be typed straight into the player to find the moment.

Speed measured 8.6x faster than realtime on a 31-minute recorded lecture, Chrome, 16GB Windows laptop using WebGPU. The limit is 2 hours per file, checked before anything starts, and it exists because the audio has to fit in the tab's memory alongside the model.

A recording a meeting app made

The apps all save a file, and all of them also offer their own transcript, which is produced by uploading the meeting to them. For a call that could not be uploaded in the first place, that is the problem rather than the solution, and the local file is the way round it.

• Zoom writes a local recording to a dated folder inside Documents, then Zoom. It contains the .mp4 and an audio_only.m4a, and the audio file gives the same transcript from a much smaller file. A cloud recording is downloaded from the Recordings page of the web portal
• Teams saves the recording to the OneDrive of the person who started it, or to the channel's SharePoint for a channel meeting. Download the .mp4 from there
• Meet saves to the organiser's Google Drive, in a folder called Meet Recordings
• A phone or a screen recorder gives you the file directly. On an iPhone it is in Photos, and AirDrop or a cable gets it to a computer unchanged

A video on a site you do not own

This is the case people most often mean by getting a transcript of a video, and it is the one this tool does not serve, on purpose. It takes a file, not a link. There is no server here to fetch anything, and a page in your browser cannot pull another site's video on its own either. A service that does take a URL is downloading that video to its own machines, which is a rights question with the video's owner and a terms question with the platform before it is a transcription question.

The platform usually has the answer already. On a YouTube video on desktop, the description's expanded view has a Show transcript button that opens the timed text in a panel, and it can be copied from there. Lecture and podcast platforms increasingly publish a transcript next to the episode.

If the video is yours and it only exists on a platform, download the original from your own creator dashboard rather than ripping the streamed copy. The original has the full-quality soundtrack and no re-encoding, and dropping it here gives a transcript the platform's auto-captions were only ever an approximation of.

Making it worth keeping

A raw transcript is a draft, and the edits that make it a record are small and specific.

• Fix the proper nouns first. They are where the errors cluster and the only ones you can find without listening again
• Switch on speaker identification if more than one person talks, then check the suggested labels. Changes sit at visible pauses on the waveform, and a line left blank is one the model was unsure about
• Keep the Markdown export. The timestamps stay as text, so a quote can always be traced back to its moment in the recording
• Use the copy-for-an-assistant export before pasting into an AI. It prefixes a note that the text is machine-transcribed, so the model hedges on a misheard name instead of treating it as fact