How to add captions to a video without uploading it

To add captions without uploading the video, the speech recognition has to run on your own computer. Most caption websites upload the file and transcribe it on their servers. Captions on CutConvert runs the speech model in your browser instead, so the video stays where it is, and so do its words. Here's how that works, what does go over the network, and how to check it yourself.

Speech recognition makes mistakes. Check names and spellings before you post.

Why it matters

• Footage you can't share. Client work under an NDA, a product that hasn't launched, an interview nobody has approved yet. Uploading it to a caption site can break the agreement you signed.
• Big files. Ten minutes of 4K from a phone can run to several gigabytes, and uploading it takes longer than captioning it.
• Nothing to delete later. A file that never left your computer isn't on anyone's server, so there's no retention policy to read.

How it works in the browser

1. The page loads Whisper, the open speech model OpenAI released, into your browser. The first time, that's a download of about 600 MB. After that it starts right away.
2. If your browser can use the graphics chip (WebGPU, in recent Chrome and Edge), it runs Whisper large-v3-turbo there, and a minute of speech takes seconds. If it can't, it runs the smaller Whisper base model on the processor, which is slower and less accurate.
3. The words come back with a time for each one, and the page groups them into captions: a few words at a time, one line or two lines, depending on the look.
4. The burned-in MP4 is encoded in the browser too, with the video encoder the browser already has. The SRT, the VTT and the projects for CapCut, Premiere Pro, DaVinci Resolve and Final Cut Pro are written on the page.

What does go over the network

• The page and the speech model, coming down to you.
• The file's name, size and length, when you ask for something that needs the unlock: the MP4 without the mark, an MP4 over 3 minutes, or a project file.
• The same three as you add a video, if you're signed in or this browser has unlocked that video before. The page checks for an earlier unlock, so you don't pay twice for one video.
• Nothing else about the video. Not the picture, not the sound, not the words.

The name, size and length are kept only for a video you unlock. That's how the unlock covers that one video, its MP4 and its projects.

Check it yourself

Open your browser's developer tools before you drop the video, and keep the Network tab in view while the captions are made. The first time, you'll see the model files come down. You won't see a request the size of your video go up. If you're signed out and take only the free downloads, you won't see its name go up either, unless this browser unlocked that video before.

Other ways to caption on your own computer

• Final Cut Pro 11 transcribes English into captions on the Mac, with Edit > Captions > Transcribe to Captions. The guide to auto captions in Final Cut Pro compares those captions with titles.
• DaVinci Resolve Studio does it with Timeline > Create Subtitles from Audio. The free version doesn't, and auto captions in Resolve without Studio covers what to do instead.
• Desktop Whisper apps, such as MacWhisper on a Mac or Subtitle Edit on Windows, run the same kind of model and write an SRT. You install them, and you style the captions somewhere else.

FAQ

Can I add captions to a video without uploading it? Yes. CutConvert's Captions page runs the speech recognition in your browser, so the video stays on your computer. Final Cut Pro's built-in transcription and desktop Whisper apps also work on your computer.

What leaves my computer? The file's name, size and length go to our server at two times only. One is when you ask for something that needs the unlock: the MP4 without the mark, an MP4 over 3 minutes, or a project file. The other is as you add a video, if you're signed in or this browser has unlocked that video before, to check for an earlier unlock. They're kept only for a video you unlock, so the unlock covers that one video. Nothing else about the video leaves your computer. The first time, your browser downloads the speech model, about 600 MB.

Why does the first run take longer? The browser downloads the speech model once, about 600 MB. After that it starts right away.

Does it work without a graphics chip the browser can use? Yes. A smaller model runs on the processor instead. It's slower and less accurate, so read the captions closely before you post.

How much does it cost? Free: the captioned MP4 up to 3 minutes with a small cutconvert.com mark in the corner, plus the SRT and VTT. One captioned video costs $5: any length, no mark, and the captions as editable text in CapCut, Premiere Pro, DaVinci Resolve and Final Cut Pro. Pro includes 10 captioned videos a month and Studio 40.