Add subtitles to a video, automatically

How auto captions work

Open your video, choose the language spoken and press Make captions. Your browser makes a sound-only copy of the video (one channel, compressed to about 0.2 MB a minute, without the file name or any tags) and sends it through uncrop.si to an outside speech AI service, which sends back the words and when each one is spoken. The captions are then built on your device: at most two lines of 42 characters each, a new caption at the end of a sentence or at a pause, each caption starting as the voice starts. Last, their words (never the sound) go through uncrop.si to an outside AI service for a clean-up that takes out filler words such as um and uh and puts back punctuation; the times stay as they are. The picture is never sent, and we keep neither the sound nor the words.

Choosing the language spoken gives the most accurate captions. Detect automatically works for most videos, and for languages that are not in the list. We tested the speech AI on read speech in English, Spanish, Portuguese, French, German, Arabic, Hindi and Japanese, and on English speech over noise and music.

Every day you get 3 minutes of auto captions free, or 10 minutes when you sign in with Google, counted by the second of sound, while the day's free pool for everyone lasts. For a longer video, open Caption only part of the video and choose the part you want; if the free time ends inside it, the captions stop at a pause so no word is cut, and the page says where they stop. Type the rest, come back after midnight UTC, or sign in for more. Typing, pasting or opening captions, every style and saving the video are always free, with no daily limit.

Speech AI can mishear names, numbers and words over music or noise, and now and then it writes words nobody said. Lines it was unsure about are marked Check this line, so read those first, and stretches with speech where no words came back are listed with Try this part again. Every line stays editable, and Undo brings back what was there before.

Only need the words? Video to text puts the transcript and its files first, and the subtitle generator goes straight to SRT and VTT subtitles.

Or add captions yourself

Type or paste the words into the box, one caption per line, or paste a whole script: a line longer than two caption lines is cut into short captions at sentences and commas, and each caption gets its line break where its two lines are closest in length. Then choose how to time them. Spread evenly shares the length of the video among the captions, longer ones getting more time, which suits a talk or a voice-over read at a steady pace. Tap along plays the video while you tap a big button, or press Space, as each caption starts; each one runs until the next, and one last tap ends the last. You can slow the video to half speed while you tap.

Already have captions? Open an SRT, VTT or TXT file. The reader is forgiving: it takes files with or without cue numbers, a comma or a dot before the milliseconds, a byte order mark, Windows or Mac line ends, UTF-8 or UTF-16 text, styling tags such as italics, and WebVTT notes and settings. Anything it cannot read is skipped and named by its line number in the file, so you can find it. A TXT file without times goes into the text box, ready to spread or tap along.

Typed, pasted and opened captions stay on your device unless you use an AI caption tool on them, cost nothing and have no daily limit.

AI tools for your captions

Once there are captions, four AI tools can work on their words: Clean up takes out filler words and puts back punctuation and capital letters; Translate turns them into one of 37 languages, every caption keeping its times; Write a post drafts a caption with hashtags for YouTube, TikTok or Instagram, and a title for YouTube; Summary and chapters writes a short summary and chapters with start times for a YouTube description (YouTube shows chapters when there are at least three, the first at 0:00, each at least 10 seconds long, YouTube Help).

They send only the words and times of your captions, never the video or its sound, through uncrop.si to an outside AI service: Clean up runs by itself right after auto captions, and every tool runs when you press it. Each press is one free AI run: 5 a day, 15 when you sign in (very long captions can count as two or more). The clean-up that runs by itself after auto captions is free. Undo brings back your captions as they were.

Burn subtitles into a video, or keep them as a file

Burned-in captions, also called open or hardcoded captions, are drawn into the picture of every frame. A caption file such as SRT or VTT stays separate, and the player draws it. This tool makes both, and they suit different places:

The burned-in video is an H.264 MP4, the kind nearly every phone and app plays, at a bitrate worked out from your original rather than a fixed default. AAC or MP3 sound, the usual kind, is copied into it as it is, with no re-encoding; other sound, such as Opus in a WebM, is converted to AAC. The caption files need no encoding, so they work in any browser, even one that cannot make the video.

Add an SRT file to an MP4

To put the subtitles of an SRT file into an MP4 so they show everywhere, open the video here, choose Open an SRT, VTT or TXT file and pick your SRT. Its captions appear in the list with their times, ready to check against the video. Pick a style, then press Save the video on the Save tab: the new MP4 has the subtitles drawn into the picture, and your original video is left as it was.

To keep the subtitles switchable instead, upload the SRT with the video where a site takes caption files. YouTube, for example, takes SRT and VTT files among its supported caption formats (YouTube Help). This tool does not add a separate subtitle track inside the MP4.

It also works the other way round: open an SRT and save it as VTT, or open a VTT and save it as SRT, with any fixes you make on the way.

Caption styles

Each style starts from its own look, and you can change its font (a clean sans, a bold display face or a serif), size, colors and position (bottom, middle or top):

Bold pop and Karaoke time each word. Auto captions carry the moment each word was heard, so these styles follow the speech. Typed and opened captions say when each line starts and ends, not each word, so each word's moment is estimated from its length within its line, and the Style tab says Estimated timing. If a word lands early or late, move the line's start or end, or split the line where the speaker pauses.

The preview is drawn by the same code that draws the saved video, and at the size the video is saved at, so line breaks, colors and positions in the file match what you saw.

Readable captions: line length, speed and timing

As you edit, lines get hints that never block anything. They follow Netflix's guide for English subtitles: at most 42 characters a line and 2 lines on screen, and a reading speed of up to 20 characters a second for adults (Netflix), with each caption on screen for at least 5/6 of a second and at most 7 seconds (Netflix). Lines that overlap in time are flagged too, because both show at once. Other languages have their own conventions; the hints use the English numbers for every language.

A line break you type is kept on screen. Captions without one wrap by themselves at the width of the picture, between words, and two lines are evened out. Chinese, Japanese and Thai are written without spaces, so there the lines break between the words your browser finds with its own word finder (Intl.Segmenter), and a word too wide for the picture breaks between characters.

Subtitles for Shorts, Reels and TikTok

For a vertical 9:16 video the tool starts on Bold pop: a few large words at a time with the spoken one highlighted, in the lower middle of the frame. Apps draw their buttons, names and text over the top, the bottom and the right of a vertical video. Meta's guide for Reels ads asks to keep the top 14%, the bottom 35% and 6% on each side free of text (Meta), so on a tall video the captions stay below 15% and above 64% of the height. Tick Show where apps put their buttons to see those zones on the preview; they are a guide, not every app's exact layout.

YouTube says Shorts can be up to 3 minutes long and can be uploaded at up to 1080p (YouTube Help), so the 1080p setting here is the most a Short can use, and 10 minutes of auto captions, the daily amount when you sign in, covers a whole Short. To save only part of a long video, cut it first; to turn a landscape video into a vertical one without cropping it, use horizontal to vertical video, then add the captions.

Any language and writing system

Auto captions understand more than 60 languages, listed in the tool in their own names. We tested English, Spanish, Portuguese, French, German, Arabic, Hindi and Japanese. Captions are drawn with your device's own fonts, so they look like the text in your other apps. A caption that starts with an Arabic or Hebrew letter is drawn right to left, with numbers and Latin names inside it kept left to right. We tested captions in English, Hindi (Devanagari), Arabic, Chinese, Japanese and Thai, and with emoji, in Chrome on Windows.

If your device has no font for Devanagari, Arabic, Thai, Chinese, Japanese or emoji (a computer without them installed, say), the page downloads a free Noto font for that writing system from uncrop.si, only then, and draws with it. The Chinese and Japanese font comes in three parts, by how common the characters are, of about 1.2 to 1.9 MB each, and only the parts your captions use are downloaded.

Which browsers can burn captions in

Burning captions in needs WebCodecs, the browser's own video encoder (caniuse), and auto captions need the browser to decode the sound in the same way. The tool checks as soon as you open a video. Where it cannot make the video, it says so and names the browsers that can, and everything else still works: auto captions where the browser decodes the sound, typing, opening and editing captions, every style and the preview, and the SRT, VTT and TXT files.

Long videos, big files and phones

The video is never uploaded, so there is no upload limit, and it is read from your device a piece at a time. Saving the video encodes every frame again, which takes time. In our tests in Chrome on a Windows laptop, a 2 minute 30 second 1080p video with 50 captions took about 35 seconds to save at 1080p and about 20 seconds at 720p, and a 20 second clip about 6 seconds. A phone can take longer. The progress bar shows the time left, and Cancel stops it at any point.

A new video of up to about 500 MB on a computer, or about 200 MB on a phone or tablet, is built in memory and then downloaded. Chrome and Edge on a computer write a bigger one straight into a file you choose. On a phone, keep the tab open and the screen on while it saves: locking the phone or switching to another app stops the save. For a long video on a phone, choose 720p, or cut the video into parts first.

Auto captions send the sound in parts of at most 1 MB, one after another, so even a long video sends little: about 0.2 MB for each minute captioned. The caption files need no encoding, so they are ready at once at any length.

What stays on your device

Your video is read by this page and never uploaded: opening it, the captions you type, paste or open, every style, the preview and saving the new video all happen in your browser.

Auto captions and the AI caption tools are the only parts that send anything. When you press Make captions, a sound-only copy of the video goes through uncrop.si to an outside speech AI service, which sends back the words, and those words (never the sound) then go to an outside AI service for a clean-up. The picture is never sent. We keep neither the sound nor the words; the speech service keeps nothing after it replies, and the clean-up service keeps requests only for a limited time for safety checks. The AI caption tools send only the words and times of your captions, when you press one of them. Under the Save tab, a line counts what this page has sent since you added your video and names the sound or words you asked to send, so you can check it yourself. The privacy page has the details.

Your captions and style are kept in this browser's storage, so you can come back to them with the same video; Start again clears them. The new video carries no location, recording date or device details from the original.