How to transcribe a video

One click turns any video into a searchable, timecoded transcript, generated privately in your browser, never uploaded to a server.

Updated July 11, 2026

Transcribing a cut usually means uploading it to some service, waiting, and hoping the audio never leaves anywhere it shouldn't. HoldFast does it the opposite way: one click, and a speech model runs in your own browser to turn the video into a searchable, timecoded transcript. The audio never touches a server.

Open the asset and generate

Open a video asset and switch to the Transcript tab. Click Generate.

Screenshot: transcribe-1
The transcript is generated privately in your browser, never uploaded.

Behind that button, a Whisper-family speech model runs locally via WebGPU. The first run downloads the model to your browser; after that it's instant. Because it runs on your machine, a capable device does the work, so generate on a laptop, then read the result anywhere.

What you get

Screenshot: transcribe-2
The result: timecoded, word-level, and searchable.

A word-level, timecoded transcript. Every phrase carries its in and out points. It's searchable, it scrolls in sync, and it becomes a second way to navigate the video: jump to a line and the player follows.

Private by design

This is a deliberate constraint, not a limitation: transcription is client-side only. Your media's audio is never sent to HoldFast's servers to be transcribed. The transcript is generated on your device and only the finished text is saved. Reading a transcript works on any device; generating one needs a browser that can run the model.

Then what?

A transcript isn't just a document. In HoldFast it's a timeline you can act on. Select a line to leave a comment or mark it up, and export the whole thing as PDF, TXT, or VTT captions. That's the next guide: Comment from the transcript & export captions.