Video to Text, Free and Without Uploading

Get text out of a video without uploading the video

Video files are the worst case for upload-based transcription. They are large, so the upload is slow and sometimes fails halfway; they are often the rawest thing you own, an unedited screen recording or an untrimmed interview; and the part that actually matters for a transcript, the audio track, is a small fraction of what you would be sending.

The tool above skips all of it. Your browser opens the container, decodes the audio track out of it on your machine, and feeds that to OpenAI's Whisper model running locally. The video itself never moves, never gets re-encoded, and never leaves your device. What comes back is timestamped text you can read, search, or export as subtitles.

How to transcribe a video file

1. Drop the MP4, MOV or WebM onto the panel above. Large files are fine, because dropping a file is not uploading it.
2. The audio track is decoded locally and downmixed for the model. The picture is ignored completely, which is why a long video is no slower than the same length of audio.
3. Let the transcript build. It streams in as it goes rather than appearing all at once at the end.
4. Export: plain text to read or search, SRT and VTT for subtitles with real timings, JSON if you want the segment data.

What people use a video transcript for

Turning a talk or a webinar into a blog post without retyping it. Making a long screen recording searchable so you can find the one minute where a decision was made. Producing subtitles so a video is watchable with the sound off, which is how most of a social feed is watched. Quoting a source accurately in an article. Studying from a recorded lecture by skimming the text instead of scrubbing the timeline.

All of those share a trait: the transcript is an intermediate step, not the deliverable. Paying per minute for an intermediate step is exactly the kind of friction that stops people from doing it at all, which is the argument for a version with no meter on it.

Frequently asked questions

Can I paste a YouTube link instead of a file?

No, and it is worth being straight about why. Pulling media down from YouTube requires a server to do the fetching, which would put a machine we control back in the middle of your files and undo the entire point of the tool. If the video is yours, YouTube Studio will hand you the original under the video's options. Otherwise use whatever local file you already have.

Which video formats work?

MP4, MOV and WebM cover almost everything people actually have, and other containers usually work too. The rule is simply whether your browser can decode the audio track, since that is all the transcriber needs. Video is never re-encoded and the picture is never touched.

Does the video get uploaded?

It does not. The browser decodes the audio track out of the container locally and the model reads that audio directly on your device. A two gigabyte screen recording produces exactly zero bytes of network traffic, which is worth knowing if you are on a metered connection or a slow hotel wifi.

Can I get subtitles rather than a block of text?

Yes. Export SRT or VTT and you get real timecoded cues, ready to import into YouTube, Premiere, Final Cut, DaVinci Resolve, or to sit next to the file for VLC to pick up. There is a dedicated walkthrough on the audio to SRT page.

My video is three hours long. Is that a problem?

Length is fine; total file size is the practical limit, at 500 MB. Long videos are often large, so if a lecture recording is over the line the usual fix is to export the audio track on its own, which is a fraction of the size and is the only part being read anyway.

Does it identify who is speaking?

No. This produces an accurate timestamped transcript, not a diarized one with speaker labels. For a two person interview the turns are usually obvious from the content and the timings, but if you specifically need automatic speaker attribution, this is not the tool for that job.