Captions that are already in time
Typing subtitles means listening to the same sentence four times and still guessing the timings. Speech recognition gives you a line per phrase with its start and end — you only have to fix the words it misheard, then choose where the text sits.
The Auto Captions Generator writes captions from the speech in a video, with a maximum number of characters per caption and a style you choose. Recognition runs on your device or on the optional server model, and the captions are styled when they are rendered on the video.
Two speakers, one timeline
Caption blocks do not sit on the video frame, they sit on time. Below, two voices share one timeline: each block is a cue, and the highlighted word is the one being spoken at that instant. The tool creates the same structure — cue by cue — and the editor lets you move a boundary without touching the text.
Three steps, no timeline work
Drop the file
The sound is extracted in the browser. With the AI server option it becomes a small mono MP3; with the on-device option nothing leaves the tab at all.
Correct the text
Each cue shows its start and end. Fix names and numbers, delete a bad line, merge what belongs together — the timings stay where they were.
Export or burn in
SRT for YouTube, VTT for the web player, or a new MP4 with the captions styled and burned into the picture.
Four caption styles
| Style | Look | Where it works |
|---|---|---|
| Bold yellow | Large, capitals, centred above the middle | Shorts, Reels, TikTok: readable with the sound off |
| Clean white | Medium outline, lower third | Interviews, tutorials, anything that has to look calm |
| Small lower third | White on a soft box, small type | Documentary and corporate footage where text is a helper |
| Karaoke highlight | Every word fills with colour as it is spoken | Music, language lessons, high-energy edits |
- check_circlePlay the first minute with the captions on
- check_circleFix names, brands and numbers
- check_circleSplit any line longer than two rows
- check_circleMake sure no cue sits under 0.8 seconds
- check_circleCheck that the last line disappears with the speech
Caption questions
Is the video uploaded?expand_more
Which languages are supported?expand_more
Can I edit the timings as well as the text?expand_more
What is the difference between SRT and VTT?expand_more
How long can the video be?expand_more
Do the captions stay in sync after burning?expand_more
Why Captions Are No Longer Optional
Most video is watched with the sound off, at least at first. In a bus, in an office, in bed next to someone asleep. Captions stopped being an accessibility feature and became the default way a large part of the audience meets a video — and platforms know it, which is why an unlabelled clip is quietly ranked lower than a labelled one.
What speech recognition actually gives you
Modern recognition models do not return a paragraph. They return phrases with a start and an end time, which is exactly what a subtitle file needs. That is the whole trick: the hard part of captioning was never the typing, it was knowing when each line appears and disappears.
What the model does badly is also predictable. Proper nouns, brand names, numbers and technical vocabulary come out phonetically. Overlapping speech collapses into one line. Music and heavy background noise produce confident nonsense. This is why the editor exists: you are not starting from zero, you are correcting a draft.
The rules that make captions readable
Two lines, no more. Around 42 characters per line. A cue that stays on screen for at least a second, and usually between two and five. A reading speed under about 21 characters per second, so the eye can follow without racing.
Break lines at natural pauses, not in the middle of a phrase. Keep the text close to the spoken words — a caption that appears a second before the voice feels like a leak from the future.
Burned in or as a file
A subtitle file is flexible: players can resize it, translate it, turn it off. Burned-in captions are part of the picture: they survive any player, any download, any re-upload, and they can be styled to match the edit. Social video usually needs them burned in, because the platform's own captions will not match your design.
The safe choice is to export both. Keep the clean master, upload the MP4 with captions where reach matters, and hand the SRT to the places that need a file.
Captioning is also an edit
Once you read your own video as text, you notice the filler. “So, basically, kind of” repeated twelve times. A sentence that goes nowhere. Captions make those visible and easy to cut, which is why editors who caption their own work tend to produce tighter videos over time.
Continue with these tools
fact_checkQuick facts
- About 15 to 17 characters per second is the speed at which a viewer can still read a caption; a longer caption leaves the screen before it is read.
- Timing matters more than styling: a plain caption at the right moment beats a styled caption that appears half a second late.
- WebVTT is the format the HTML video element reads for captions, and it separates the millisecond with a dot.menu_bookW3C
- The older SRT format has no formal standard; it survives because every player imports it.menu_bookW3C
- A language model reads text in tokens, and one token is roughly four characters of English, so a 300-word answer is about 400 tokens.menu_bookOpenAI Cookbook
helpQuestions people ask
How do I add automatic captions to a video?expand_more
Are automatic captions accurate enough for YouTube or social media?expand_more
How long should one caption stay on screen?expand_more
hubWhat this page is about
- WebVTTopen_in_new
The W3C subtitle and caption format used by the HTML video element, with cue timings written as 00:00:01.000.
- Large language modelopen_in_new
A model that predicts the next token of a text; the site sends the prompt to a hosted model and returns its answer.
- H.264 / AVCopen_in_new
The video codec standardised by ITU-T as H.264 and by ISO/IEC as MPEG-4 AVC, decodeable by every current browser.
library_booksOfficial sources cited on this page
Continue with the same file
Each of these opens what the tool above produces. Use “Use in another tool” under the result to carry the file over — it moves from tab to tab inside your browser, never to a server.
Switched-off tools are not listed here: only what works right now.