Does my video leave my computer?
No. Your browser decodes the audio, runs the Whisper speech model, and encodes the captioned video. Our server only answers the payment check.
Which model is it?
Whisper base, converted to run in browsers with transformers.js. It uses WebGPU when your browser has it and WebAssembly when it does not.
Why is the first run slow?
The model files, about 80 MB, download once and your browser caches them. After that a clip starts right away.
How accurate is it?
Good on clear speech in major languages. Weaker on heavy accents, music under speech, overlapping voices and rare names. That is why every line is editable.
Which languages?
Around 99. Pick one, or leave auto detect on. Tick translate to get English text from any other language.
How long can a video be?
3 minutes free, 30 minutes on a paid plan.
Will the burned-in video look the same everywhere?
Yes. The font is bundled and the text is drawn into the frames, so every player shows it identically.
Why does the burn step take a while?
ffmpeg runs in a single thread inside the tab. Plan on one to three times the clip length. Keep the tab in front.
Do I need an account?
No. A paid plan is tied to the browser you paid in, through a cookie. Clearing cookies removes access on that browser. Reply to your Stripe receipt email and we will restore it.