This page won't work opened directly as a file.
Browsers block Web Workers (needed for model loading/transcription) when a page is opened via file://. Serve this folder over local HTTP instead, e.g.:
npx serve .
# or
python -m http.server 8000Then open http://localhost:<port> in your browser.
Add subtitles to an MP4 entirely in your browser — transcribe with Whisper, edit the text, translate it, then download SRT/VTT or burn the subtitles into the video. No upload, no server, no sign-up.
For a non-English language its recommended you set the language yourself.
The model you choose is downloaded once and then cached by your browser, so picking it again later is instant. Switching to a different size downloads that one the first time you use it.
Drag & drop an MP4 file here
or click to browse
Re-encodes the video with subtitles permanently drawn into the frames (hardsub). Runs locally via ffmpeg.wasm; can take a while for longer videos.
Translates the text currently in the box above (including your edits), replacing it in place. Requires "Spoken language" above to be set to a specific language, not Auto-detect.
Local Whisper Subtitles is a free subtitle generator that runs completely on your own device. It uses OpenAI’s Whisper speech-recognition model, compiled to WebAssembly and executed through Transformers.js, to turn the speech in a video into timed subtitles. Burning subtitles into the picture is handled by ffmpeg.wasm. Because both run in the browser, your video is never uploaded, there is no account or API key, and there are no file-size or minute quotas beyond what your own hardware can handle.
.srt or .vtt file, or burn the subtitles permanently into a new MP4.No. Transcription, translation and re-encoding all happen inside your browser using WebAssembly and, where available, WebGPU. There is no backend to send a file to.
SubRip (.srt) and WebVTT (.vtt). You can also burn the subtitles into the MP4 as hardsubs so they show in any player, including ones that ignore sidecar subtitle files.
Whisper can auto-detect the spoken language, and English, Arabic, Spanish, French, German, Hindi, Chinese, Russian, Portuguese and Turkish can be selected explicitly. Finished subtitles can be translated to English or Swedish.
Tiny (~75MB) is the fastest and leans English-only. Base (~145MB) is a good balance and handles non-English audio well. Small (~480MB) is the most accurate but noticeably slower. You pick which one to download — the page never fetches a model on its own.
Yes — completely free, with no account, API key or usage cap, because all the computation runs on your own machine.
Three referenced explainers on the machinery behind this tool — speech recognition, machine translation and the subtitle formats themselves. Every claim links to a primary source.
The six stages between a video file and a translated caption track — audio extraction, speech recognition, cue segmentation, translation, re-timing, rendering — and how to tell which one caused a given mistake.
Log-Mel spectrograms, the encoder–decoder transformer, the special task tokens that make transcription and translation the same model, why it sometimes hallucinates text over silence, and what really differs between tiny, base and small.
Why an MP4 is not a codec, the difference between soft and hard subtitles, SRT versus WebVTT in detail, the 23.976 fps sync problem, professional reading-speed limits, and what burning in subtitles actually costs.