Transcribe and caption video on your own machine
Drop in a lecture, a meeting or a clip. Whisper runs in your browser, so the file never gets uploaded anywhere. No account, no minutes cap, and up to two hours per recording.
- No upload
- No signup
- No watermark
- No file size limit
Drop a file
Audio or video. Open the network tab if you want to check nothing leaves.
Already have captions? Import an .srt or .vtt after you drop the file.
Read straight off your disk. Works offline once the model is cached.
How it works
Four steps, all of them on your own machine.
Drop a file
Audio or video, read straight off your disk. There is no upload bar because nothing is uploaded.
Transcribe
Whisper runs on your GPU, or your CPU if the browser has no WebGPU. The transcript fills in as it goes and you can stop it partway.
Fix what it got wrong
Edit against the waveform. Lines too fast to read, cues that flash by, and the spots where the model looped all get flagged.
Export
Markdown or plain text for a transcript. For a video, an MP4 with the captions rendered in, or a plain .srt.
Nothing gets uploaded
Whisper and the video encoder both run in your browser. There is no upload step because there is no server to upload to. That is the part that makes it usable for a client call or a consultation you are not allowed to put through a web service.
Transcript or burned-in captions
Same engine, two outputs. A timestamped transcript as Markdown or plain text, or an MP4 with the words rendered into the picture. Most free caption tools hand you an .srt and charge for the second one.
Nothing to meter, no account, no watermark
Your machine does the work, so a minute of audio costs us nothing. No 300-minutes-a-month tier, no file size limit, and no reason to stamp our name on your video. The one real limit is length: two hours per recording, because it all has to fit in a browser tab.
Where the free tools stop
These are the walls you hit after you have already waited through the transcription.
| Typical free tool | CaptionSuite | |
|---|---|---|
| Burning captions in | Paid plan only | Included |
| Watermark | On the free export | Never |
| Account | Sign up first | Not needed |
| Minutes cap | Free tier runs out | Nothing to meter |
| File size | Capped, often at 1GB | No limit |
| Your video | Uploaded to their servers | Never leaves your device |
How it works
All of it happens in the tab. Your browser decodes the audio with its own video decoder, resamples it to 16kHz mono, and runs Whisper through WebGPU. Burning captions in goes the other way: decode each frame, draw the text onto it with Canvas, re-encode with WebCodecs.
One thing does get downloaded, once: the model weights from Hugging Face, 45MB or 80MB depending on the size you pick. Your browser caches them after that, so the second file you transcribe starts straight away. That is a download to you. Nothing goes the other way, which is also why there is no usage limit to enforce.
Why burning in is the bit that gets paywalled
Most free caption tools generate an .srt and then ask for money to put the words on the video. For anything going to a social feed that is the wrong half: Reels, TikTok and Shorts take a video file and no sidecar, so the .srt is the part you cannot use.
It gets paywalled because on a server it genuinely costs money. Re-encoding video is real compute per user. Here the encoder is yours, so there is nothing to charge for.
If you are not allowed to upload it
Legal recordings, consultations, HR interviews, internal calls. The material that most needs transcribing is usually the material you cannot send to a web service. There is no server here, so there is nothing to send. Check it in the network tab: the model comes down, nothing goes up.
What it is bad at
Speed depends on your hardware, not a data centre. With WebGPU, in Chrome or Edge on a recent desktop, transcription runs about 8x faster than realtime. Without it you are on a single CPU thread and it is much slower. On a phone short clips are fine, but it throttles as the device warms up.
There is also a length limit, and it is worth knowing before you start. Everything happens in one browser tab, and a tab has a memory budget: two hours per recording to transcribe, and about half an hour of 1080p video to burn captions into, because the finished MP4 has to exist in memory in one piece before your browser can save it. File size is not limited at all, and neither is how much you run through it. The numbers, and where they come from.
Accuracy is decent, not perfect. It gets proper nouns, acronyms and technical vocabulary wrong, and it loops over music and long silences. The editor flags the lines most likely to be wrong rather than pretending the transcript is finished. There are no speaker labels.