How to add captions to a video without uploading it

It means the video file itself, the frames and the original audio, stays on your device. It does not mean that nothing is sent. To transcribe speech, CutXflow has to send the speech somewhere, so it sends a small speech-only audio track. Everything else, including drawing the captions and encoding the MP4, happens in your browser.

The audio goes to CutXflow's server, which checks that you are signed in, forwards it to ElevenLabs's speech-to-text service, and returns the transcript. The transcription route does not write the audio to a database or to file storage. It logs the language, length, word count and credits used, not the audio and not the words. CutXflow's privacy policy says CutXflow does not save the audio and that ElevenLabs may keep a copy under its own policy.

What happens to the audio at ElevenLabs is governed by ElevenLabs's own terms and account settings, not by CutXflow. Its documentation says customers by default have history preservation enabled, and describes an optional Zero Retention Mode, offered to enterprise customers, in which speech-to-text audio and text are deleted as soon as the request completes (ElevenLabs: Zero Retention Mode (opens in a new tab)). So the accurate short statement is that the audio is sent to a third party for transcription. How long that provider keeps it is set out in its documentation and in CutXflow's privacy policy; read both if it matters for your footage.

Transcription runs on a hosted speech-to-text model, not inside your browser. CutXflow cannot produce word-level Hindi transcripts without sending audio to that service. If a clip cannot leave your device in any form, CutXflow is the wrong tool for it, and a fully offline workflow is the right one.

• Reading. The browser reads the file you chose into the tab's memory. No request is made for it.
• Extracting the audio. CutXflow uses the browser's own WebCodecs (opens in a new tab) where it is available, and otherwise an audio engine compiled to WebAssembly, about 31 MB, fetched from CutXflow's own site. Either way it runs on your device. The result is 16 kHz mono Opus at 32 kbps.
• Drawing and encoding. The captions are drawn onto each frame in the browser, the frames are encoded by the browser, and the finished file is saved from memory to your Downloads folder. During export the only requests the app makes are account and billing checks and loading the watermark image from its own site. No video data is involved.

You can check this yourself. Open your browser's developer tools, go to the Network tab and caption a clip. The one large upload is the audio request to /api/transcribe, and the editor reports how many kilobytes of audio were sent once transcription finishes.

It is a good fit if the thing you want to avoid is sending a whole video file to a third party, and you are comfortable with the speech in it being processed by a transcription service under CutXflow's privacy policy. The audio track is a small fraction of the size of the video, and the picture never leaves your machine.

According to the privacy policy, CutXflow stores:

• Your email address and, for Google sign-in, your display name, through Firebase Authentication.
• If you are signed in, your project data: caption text, timings and style settings, as in the table above.
• Payment confirmation and plan status from Cashfree, never your card, UPI or bank details.
• Standard server logs, including IP address, for rate limiting and abuse prevention.

The policy also says CutXflow does not sell personal data and does not use your content to train machine-learning models. For the captioning steps themselves, see how to add Hindi captions to Reels and Shorts.