Auto Caption Video

Turn the speech in a video into editable captions on your device, then download SRT or VTT subtitles, burn them into the picture or add a subtitle track, without uploading the video.

Choose a video or audio file

or drop it here

The video is processed on this device and never uploaded.

MP4, MOV, WebM, MKV, AVI, WMV, FLV, 3GP, TS / MTS / M2TS or VOB / MPGAudio: WAV, MP3, M4A, FLAC or OGG

Files over 1 GB can be slow. Very large outputs need a browser that can save straight to a file.

How it works

Choose a video and press Transcribe. The page reads the sound track, finds the parts with speech and runs OpenAI’s Whisper speech recognition model on them in your browser. Each word gets a time, and the words are grouped into captions of one or two short lines.

The model files download once, from this site. Whisper tiny is 44 MB, base 80 MB and small 252 MB when the model runs on the processor (WebAssembly), or 122, 209 and 589 MB on the graphics card (WebGPU), plus 29 MB for the speech engine. Your browser keeps them, so later transcriptions work offline. The video and the transcript never leave your device.

Read the captions and fix anything the model misheard; changing the words keeps their timing. Then download an SRT or VTT subtitle file, burn the captions into the picture so they always show, or add a WebVTT subtitle track that viewers can switch on and off. Burning in re-encodes the video. The subtitle track copies the video and audio without re-encoding when the format allows it.

With WebGPU the model runs on the graphics card and is several times faster; otherwise it runs on the processor.

Supported files

Inputs: MP4, MOV, M4V, WebM and MKV. Transcription also works for AVI, WMV, FLV and MPEG files, for audio files (WAV, MP3, M4A, FLAC, OGG) and for audio the browser cannot decode itself (WMA, Apple Lossless, AMR, Dolby TrueHD and others): the FFmpeg engine, downloaded once from this site, decodes the audio on your device first.
Outputs: SRT and WebVTT subtitle files; MP4 (H.264) or WebM (VP9) with burned-in captions; MP4, WebM or MKV with a WebVTT subtitle track.

Limitations

Accuracy depends on the recording. Clear speech close to the microphone works best; music, crosstalk, strong accents and noise cause mistakes. Read the captions before you publish them.
Whisper understands about 100 languages, but quality varies widely. English, Spanish, French, German, Italian, Portuguese and Japanese are the strongest. Automatic detection uses the first speech in the video.
The first transcription downloads the model: 44 to 589 MB, depending on the model and on whether the browser has WebGPU. Small is offered only where this site hosts it.
Speakers are not labelled, and sounds or music are not described.
Long videos take time. On a typical laptop without WebGPU, the base model needs roughly as long as the speech itself.
Browsers without WebCodecs audio decoding, such as Firefox for Android, decode the audio with the FFmpeg engine first, which is slower and needs one online visit to download it.
Burning captions in or adding a subtitle track to WMV, MPEG-2 (VOB) or MPEG-4 Part 2 video, which browsers cannot decode, first decodes the video into an intermediate copy with the FFmpeg engine, which takes longer. Audio files get SRT and VTT captions only.
Burning captions in re-encodes the video. Web browsers do not show a WebVTT track stored inside a file, many players and editors ignore it, and MOV files cannot carry one, so they are saved as MP4.

Questions

Is my video or transcript uploaded?

No. The speech model runs in your browser, and the video and the transcript never leave your device. The model files download once from this site and stay in your browser.

How large is the model download?

Whisper base, the default, is 80 MB on the processor or 209 MB with WebGPU. Tiny is 44 or 122 MB and small 252 or 589 MB. The speech engine adds 29 MB once. Later visits reuse the stored files.

Which model should I choose?

Base is the best balance for most videos. Tiny is faster and smaller for older devices but makes more mistakes. Small is the most accurate and needs a recent computer.

Should I use SRT, VTT, burned-in captions or a subtitle track?

Upload an SRT or VTT file to YouTube, Vimeo or a course platform. Burn the captions in for social media, where subtitle files are often not supported. Add a subtitle track when you share the file itself and viewers should be able to turn the captions off.

Can I edit the timing?

Yes. Each caption has start and end times you can change, and you can split, merge, delete or add captions. Tap or click a caption’s time to play the video from there.

Example

A two-minute interview recorded on a phone: choose it, keep the base model and automatic language, and press Transcribe. After the model has downloaded once, the captions appear in about a minute; fix the names it misheard, then press Burn captions into video to get an MP4 for social media, or Download SRT to upload with the original.