Auto-Generate Subtitles from Video

Yes. No sign-up and no limits.

No. The AI runs inside your browser, so your video and audio are never sent anywhere. Only the AI program files are downloaded from a public repository the first time.

The first time, your browser downloads the AI model: on a regular computer (CPU) about 80 MB for Fast and 240 MB for Accurate. Browsers that can use your graphics card run much faster but use somewhat bigger files (Fast about 160–200 MB, Accurate about 290 MB, Most accurate about 540 MB). After that it isn't downloaded again in the same browser.

It depends on the model and the audio. Clear English speech usually comes out well even with Fast. For Korean we measured on 15 public speech clips (about 2 min 40 s): about 15% of characters were wrong with Fast, about 7% with Accurate and about 3% with Most accurate. Those were clearly read sentences; real videos with noise or fast speech will have more errors. Names, jargon and numbers are the most common mistakes, so skim the text before saving.

Yes, but speed varies a lot between computers. On our test PC an 11-minute video took about 1.5 minutes with Accurate on the graphics card, and about 4 minutes with Fast on the CPU alone. Accurate on the CPU can take about as long as the video itself. The whole soundtrack is held in memory, so videos over an hour or files over 1 GB may make the browser stall or fail — extract the audio with 'Video to MP3' first or split the video.