How to add subtitles to a video

To add subtitles to a video you need three things: the words that are spoken, a time for each line, and a way to put them on screen. You can type them by hand, but it is far quicker to let a speech model write them, check them, and then either burn them into the picture or save a subtitle file such as SRT. The steps below do all of that in your browser with Inky Clips, free, and without uploading the video anywhere.

Why add subtitles without uploading

Almost every auto-caption tool works the same way: you upload your video, a server transcribes it, and you download the result. That’s fine for a cat video. It’s less fine for a client’s unreleased product, a patient interview, a lesson with children in it, or anything under an NDA. Once the file is on someone else’s server you are relying on their retention policy.

There’s another way now. Speech recognition models have become efficient enough to run in a web browser, on your own computer’s processor and graphics card. Inky Clips uses this to caption video without the video ever leaving your machine.

How it works

The captions come from Whisper, the open speech recognition model released by OpenAI, in its large-v3-turbo version. It understands 99 languages and gives a timestamp for every word. The editor runs it inside your browser, in a background worker:

• Your video is opened from your disk by the browser, the way a photo viewer opens a file. It isn’t sent anywhere.
• The sound is decoded in the page and handed to the speech model running in the same tab.
• The model downloads once. The first time you caption, the browser fetches the model files from Hugging Face, where they’re published. That’s 1.1 GB, or about 1.6 GB on computers that can use the faster graphics-card version. The browser keeps them, so later videos skip the download.
• The text is drawn into the video on your computer when you export, and the MP4 is saved straight to your downloads.

Our server is involved in exactly one thing: Google sign-in. Your projects are kept in your browser’s storage on this computer, not in an account. The privacy policy spells this out.

How long it takes

Speed depends on your hardware. The editor uses your graphics card through WebGPU when the browser and GPU support it, and falls back to the processor otherwise. The graphics card route is usually much faster; on the processor alone, long videos are worth leaving to run while you do something else. The dialog shows how far it has got, and you can cancel at any point.

A few practical points:

• Keep the tab open during the first download and while it listens. Closing it stops the work.
• The model needs memory. If a large video fails on an older machine, close other heavy tabs and try again.
• A video without a sound track can’t be captioned, and one with only music may come back with no speech found.

After the captions appear

The transcript is split into short captions of up to six words, breaking where the speaker pauses. They show on their own purple lane in the timeline. Click any one to correct its words or timing. Right-click (or long-press) a caption and choose Edit all captions to change the font, size, colour or position of all of them at once. Word-by-word captions covers editing in detail.

If you run Auto captions again on the same video, for example with a different language, the new captions replace the old ones and keep the style you set.

What you get, and what you don’t

The captions are burned into the picture: drawn onto every frame, in a handwriting or subtitle font of your choice, writing themselves as the words are spoken. That means they look identical everywhere, including platforms that ignore subtitle files, but viewers can’t switch them off. If you need closed captions as well, for YouTube or a broadcast, choose Export subtitles in the editor and save an SRT or VTT file (or a plain text transcript) made from the same captions, also without uploading anything.

Languages

You pick the spoken language from a searchable list of all 99; the editor remembers your last choice. English, Spanish, German and other widely spoken languages come out cleanest. Smaller languages work too, with more words to check. See auto captions in Latvian and other smaller languages and the full list of supported languages.