Burncaps burns animated word-by-word captions into a video, or cuts a talking video to its strongest 30 or 60 seconds, entirely inside your browser. It is made by Jordan Labs Ltd, a company registered in England and Wales, company number 17358646. Support and refunds go through one address that a person reads: hello@burncaps.app.
The heavy work happens on your own machine. The speech model transcribes locally and your browser's own video encoder writes the finished MP4, so no video ever touches a server we pay for. The one exception is the sound, and only when you choose it: a phone, or a video in an Indian language, can use our speech service, which costs us about a cent a video, and a fair monthly allowance keeps it that way. There is no per-video cost worth charging a subscription for, which is why Pro is $49 once rather than a monthly bill, and why you can try everything first: every feature comes free, with no account and no watermark, for two videos every 30 days. Tools that upload your file have to charge you every month for the servers doing the work; we do not have that bill. Subtitle files (SRT, VTT, ASS, text) are free without limit for recordings up to ten minutes, and for the first ten minutes of a longer one.
Do not take the privacy claim on faith, it takes about thirty seconds to test. Open your browser's developer tools, select the Network tab, then caption a video here. The first run downloads the AI model once from Hugging Face's CDN (roughly 190 MB with a graphics chip, 80 MB without) and caches it; a video in one of twelve Indian languages also downloads the model made for that language, 138 MB once, from the same CDN. The caption fonts come from this site itself; the person-cutout and face models, the demo video at the top of the home page, and the code that runs the speech model, the person models and the video writer come from jsDelivr, a public CDN; and each page asks our server once which price applies where you are, a request that carries nothing about you. You will also see small step counts going to /api/count (the step's name, where the visit came from, phone or computer and the country, never an identifier) and Cloudflare's page counter, which cannot identify you. The face-detection library tries to send usage statistics to Google every minute; this site's security policy refuses that, so it shows as blocked in the console. After that you will see no request carrying your video, because there is nowhere for it to go, and no request carrying your sound unless you chose our speech service: then you will see pieces of sound going to /api/listen and the words coming back. Transcribing, correcting and previewing keep working with your connection off. Exporting needs it for one small check: before the first frame is written, the page asks our server for an export pass, sending a random ID for your browser and a random ID for the file, never anything from the video itself, and when the file is finished it says so with the same IDs: only a finished export uses a free video. That is how the two free videos are counted fairly and Pro stays for the people who bought it. One more request, and it is opt-out: when you choose the best 30 or 60 seconds, type names to spell right, or choose a caption language other than the one spoken, the transcript text (the words, with their timings for the pick; the lines alone for a translation) is sent to our AI service to choose the strongest stretch, fix the spelling of names, or translate the lines. It is not stored. Tick "Keep everything on this device" and that request never happens either. Leaving your email on the Studio or Clips card sends that email, and nothing else from the page, to /api/interest; asking for our emails after an export sends it to /api/news, which first sends one email to confirm the address is yours and nothing else until you do, and emailing yourself this page from a phone sends it to /api/send-link, which keeps nothing.
Pro is $49 once in most countries (₹998 in India, by UPI or card, and $10.40 in lower-income countries, the same amount), not a subscription, and it buys one thing: unlimited videos with no mark, either output, with every style and the maximum-accuracy model included as they are in the free videos (Pro also includes unlimited translations and best-moment picks, with names checked, and the whole script and subtitle files of a recording up to 90 minutes). Card payments are handled by Stripe, so we never see your card details. Pro switches on the moment the payment clears. There is no key to keep: Pro is tied to the email you pay with, and on another device you sign in with that email and a six-digit code we send to it. If Pro does not work on your machine, email us within 14 days and we refund in full.
As many as you like. There is no daily, monthly or lifetime limit on videos or exports, and none on subtitle files (SRT, VTT, ASS and text), which are free for everyone for recordings up to ten minutes; with Pro they cover the whole of a recording up to 90 minutes. No video carries a mark.
Video with captions burned in, and the best 30 or 60 seconds: up to ten minutes and 1.2 GB per file on a computer, and 250 MB on a phone. Pro and free are the same here. The reason is that your browser holds the file in memory while it writes the new video, and a longer recording is more than many computers can hold at once. A computer writes the whole video at its own resolution. We have not measured a ten-minute 4K export, so we do not promise one: try a 4K clip of yours on the free videos first. A longer recording, up to 90 minutes and 10 GB on a computer, opens for its script and subtitle files, because only its sound is read, a piece at a time: without Pro the first ten minutes are free, and Pro does the whole recording. We have no date for burned-in video past ten minutes.
One request is one batch of your caption lines sent to our AI. A video of normal length is one or two requests, and each line you correct afterwards is one more. With Pro, translations and best-moment picks are unlimited for your own use. Safety limits per account, for a day and for any 30 days, far above what one person making videos reaches, exist only to stop a misused sign-in, and we do not publish the numbers. The daily one resets at midnight UK time. If real work ever reaches it, the page says so, the captions stay in the spoken language (or a local rule picks the stretch), and an email to hello@burncaps.app gets it lifted.
The AI answers in short pieces of two to four spoken words, in the order they were spoken, and each piece is shown over the time of the words it translates. So the English follows the speaker, instead of being spread evenly across a sentence. Any line can be corrected by hand.
Five hours of sound in any 30 days, against 25 minutes without Pro. That is where a phone, an Indian language or any non-English language is written in seconds, by Microsoft’s MAI-Transcribe. Forty videos of five minutes are about three hours twenty. Only listening on our service uses it. When the five hours are used, Burncaps listens on your own device instead, without a limit and more slowly on a phone, and says so.
Pro is tied to the email you paid with and can be signed in on three devices at once, for example a phone and two computers. To add a new computer, choose Pro sign-in on it, type that email, then type the six-digit code we send to it. There is no password. A fourth sign-in signs out the device used longest ago, which can sign back in the same way. Please do not share your sign-in with other people.
Pro is $49 once, and features added to Burncaps later come with it at no charge. There is nothing to renew, because Pro is not a subscription. A separate plan for teams and client work, called Studio, is only being asked about; if it is ever built it will be a different product with its own price. Nothing that is in Pro today will be moved out of it, into Studio or into any other plan. You can have your money back within 14 days if Pro does not work on your machine.
Your video never leaves your device. A video you drop on one of our other pages waits in your browser’s own storage on your device until the editor opens it. The sound leaves only if you choose our speech service, in pieces of up to 30 seconds with nothing from the picture, to Microsoft’s MAI-Transcribe in Ireland; neither we nor, by Microsoft’s statement, Microsoft keep it. The transcript text goes to our AI only for a best-moment pick, a name check or a translation, and is not stored. What we keep is counts, stored as salted one-way hashes, and for Pro your email, your purchase and up to three signed-in devices. Ticking “Keep everything on this device” stops every request that carries sound or words. The privacy policy has the whole list.
Burning captions into video uses WebCodecs, which today means a Chromium browser on a desktop or laptop: Chrome, Edge, Brave or Arc. In Safari and Firefox you can still transcribe and export subtitle files, but not the finished video. Input files are MP4 and MOV. Long recordings take longer on slower machines because the work is yours, not a queue on our side. AI transcription makes mistakes, so every word stays editable until you export.
OpenAI's Whisper speech model in its open-source MIT-licensed ONNX builds, run through transformers.js; for Hindi, Urdu, Bengali, Tamil, Telugu, Gujarati, Marathi, Kannada, Malayalam, Punjabi, Nepali and Odia, AI4Bharat's IndicConformer speech models (MIT licence, in the ONNX builds published by OpenVoiceOS), run through ONNX Runtime Web; the browser's WebCodecs encoder with mp4box and mp4-muxer for the export. Burncaps is not affiliated with OpenAI, AI4Bharat, Submagic, CapCut, TikTok, Instagram or YouTube.
Email hello@burncaps.app for support, refunds, bugs or anything about your data. There is no ticket portal and no bot to get past. Legal detail lives in the terms and the privacy policy.