Subtitle Generator — Auto-Generate Subtitles from Video or Audio (Free)

Subtitle vs caption — which do you actually want?

Most tools use the terms interchangeably. The technical difference matters for accessibility compliance.

For the full breakdown of ADA, WCAG 2.1 SC 1.2.2/1.2.4, and the Title II web rule, see What is closed captioning?

How the subtitle generator works

Which format do I need for my platform?

Every platform takes subtitles differently, and two of them will not take a file from you at all. Where a platform's own help page is years out of date or does not answer the question, this table says so rather than guessing — which is why a couple of rows read “not documented”.

For a deeper dive on the SRT format, see what is an SRT file? For YouTube specifically, YouTube Studio SRT upload is documented at support.google.com/youtube. Already have an SRT and just need to translate it to another language? Use the dedicated subtitle translator — timestamps preserved, 133 target languages.

Language coverage and accuracy

Accuracy varies by language, by audio quality, and by how many people are talking at once — far more than it varies between engines. We do not publish an accuracy percentage for this tool, because a single number measured on clean benchmark audio tells you almost nothing about your recording.

What reliably costs you accuracy: background music under speech, two people talking over each other, heavy accents in a language the model saw little of, and technical vocabulary or proper nouns that were never in the training data. What reliably helps: a clean single-speaker recording at a normal pace.

The practical answer is to run a minute of your own audio through it and read the result. That takes less time than reading anyone's benchmark table, and it is the only test that measures the thing you actually care about. If you do want published figures, we keep the independent ones on how accurate is Whisper — with the caveat that those benchmarks measure clean read speech, not your conference call.

Burned in, or a separate file?

A subtitle file sits alongside the video and the player draws it. Burned-in captions are part of the picture, rendered into the frames before you upload. Most advice on this choice repeats three claims about the second option, and none of them survives a look at the source.

The honest case for captions is simpler and better evidenced. In AP-NORC polling from August 2025 (1,182 US adults, ±3.8 points), a third of people use subtitles always or often, and it skews young. The reason most of them give is not a muted phone: half say they want to catch every word, and four in ten are following an accent or a second language.

So the real decision is not about accessibility or search. It is about control. A separate file can be switched off, restyled by the viewer, translated into another language without touching the video, and corrected after publishing. Burned-in text can do none of those — but it always renders, identically, everywhere, which is worth a great deal on platforms whose caption support is patchy.

In practice the platform usually decides for you. Where a subtitle file is accepted, upload one. Where it is not — Instagram's organic posts being the notable case — burning in is the only route, and no accessibility standard holds that against you.

Either way the first step is the same: generate the subtitle file here. Burning it in is a separate pass in a video editor or with ffmpeg, and our guide to adding an SRT to an MP4 has the command.

Subtitles for vertical video are a different shape

Almost every subtitle rule you will read was written for landscape television. A 9:16 frame changes the arithmetic, and the BBC is one of the few bodies to have written the difference down — its guidelines added vertical-specific guidance in a recent revision, and Netflix's style guide still does not mention 9:16 at all.

The first two rows are the same rule seen from two sides. A vertical frame is 1080 pixels wide where a landscape one is 1920, so even with a more generous width allowance each line holds roughly a third fewer characters. The extra line is not permission to say more — it is compensation for lines that hold less.

If you are starting from a video file rather than a subtitle file, video to SRT covers that route and the cue-splitting rules behind it.

One more piece of vertical-specific advice worth knowing: the BBC suggests placing subtitles a little higher than you would in landscape, because in a vertical frame faces tend to sit in the top half and the usual bottom position covers more of what matters.

Figures from the BBC Subtitle Guidelines (v1.2.5, March 2026), sections 3.1, 3.3, 9.2, 10 and 10.1, read on 11 September 2026. They are written as recommendations rather than hard limits, and the document allows an extra line where you are confident nothing important is covered.

Before you upload the file

Rejections are rarely about the subtitles themselves. They are almost always one of these, and each is documented by the platform that enforces it.

• Save as UTF-8. YouTube requires it. A file saved as Windows-1252 turns accented characters into mojibake, and the damage is invisible until it is on screen.
• Number the cues from 1, without gaps. Facebook documents this explicitly, and a renumbering slip is the usual cause of a file that imports with cues missing.
• No trailing blank line. Also a documented Facebook rule, and the reason a file that looks fine in a text editor can still be refused.
• Watch the size for Facebook. Their caption files are capped at 200 KB, and the filename has to carry the locale:video.en_US.srt.
• Strip styling markup for YouTube. They accept the basic version of each format and ignore markup, so font and positioning tags are wasted at best.

Our free SRT editor checks all of these in a browser tab: it renumbers on export, writes UTF-8 without a byte-order mark, strips markup on request, and flags cues that run too fast or too long before a platform ever sees them.

Editing subtitles after generation

No account? The free SRT editor opens the downloaded file in a browser tab and flags cues that read too fast or run too long — no upload, nothing to sign up for. With an account, every subtitle project also opens in a full editor with the video preview synced to the cues, where you can:

• • Fix misheard words with click-to-edit — timestamps update automatically.
• • Merge or split cues at word boundaries for readability.
• • Rename speakers and preserve speaker labels through SRT export.
• • Adjust cue duration to match speaker delivery (dramatic pauses, fast dialogue).
• • Add non-speech notation like [applause] or [music] for accessibility compliance.
• • Re-export as SRT, VTT, TXT, or a re-burned MP4 without regenerating from scratch.

Bulk generation

Upload up to 50 files at once for batch processing on any paid plan. Useful for course creators subtitling a full module, agencies processing a client backlog, or podcasters generating show-note transcripts across an entire season. Each file lands in your library as an independent project with its own editor and export options.

When to hire a human transcriptionist instead

AI subtitle generators land at 92–96% accuracy on clean audio. For contexts where the remaining 4–8% is unacceptable, use a human service:

• • Broadcast television — FCC accuracy standards require near-perfect captions.
• • Legal proceedings — court transcripts and depositions need certified accuracy.
• • Medical dictation — drug names and procedures where a single misheard word has clinical consequences.
• • Named entities under legal review — patents, contracts, regulatory filings.

A common workflow: run AI generation first (fast + cheap), then hand the SRT to a human editor for a review pass. Services like Rev and Happy Scribe offer human review at $1.50–$2.00 per audio minute — see our comparison of subtitle tools for details.