Add Subtitles to a Video Without Uploading It Anywhere

Most people watch video with the sound off. Scrolling in a lecture hall, on a train, next to a sleeping child, in an office — the audio is a luxury and the captions are the content. A video without subtitles is a video most of its audience never actually hears.

The usual fix is to hand the file to a website, wait, and hope. That means your unpublished footage, your client's screen recording, or a video of your family sits on someone else's server for a while. For a cat video, fine. For anything you have not published yet, it is a strange thing to agree to just to get some text on screen.

There is now a third option. Speech recognition models small enough to run inside a browser tab have existed for a couple of years, and they are good. Your laptop does the listening.

What "on-device speech recognition" actually means

Auto Subtitle Generator works in three steps, none of which involve a server:

1. The model downloads to you. The first run fetches the speech-recognition model — a one-time download, cached afterwards. This is the part people mistake for an upload. It is the opposite: the model is travelling to your device, not your video travelling away from it.
2. Your browser extracts the audio from the video file you picked, entirely locally.
3. The model transcribes it, with timings, and you get subtitles.

The proof is simple and you can run it yourself: load the page, turn off your Wi-Fi, and generate subtitles anyway. It works. Nothing was uploaded because there was nothing to upload to.

If it is only audio you have — an interview recording, a voice note, a podcast — Speech to Text runs the same class of local model and exports TXT, SRT or VTT.

SRT file or burned-in? This is the decision that matters

Everything downstream depends on this, and most guides skip it.

A subtitle file (.srt or .vtt) is a separate text file listing each line and its timecodes. The video is untouched. YouTube, Vimeo and most players accept one, and the viewer can turn captions on or off. It is also editable forever — fix a typo in a text editor, no re-encoding.

Burned-in subtitles are drawn permanently into the video frames. There is no off switch and no editing afterwards. That sounds worse, and for YouTube it is. But Instagram, TikTok, WhatsApp status and LinkedIn's feed either ignore subtitle files or bury them, so burned-in is the only version most viewers will ever see.

The rule of thumb:

• Uploading to YouTube or a player you control? Export the SRT. Keep it flexible.
• Posting to social? Burn them in. Subtitle Hardburn does it with a live styled preview so you can see the result before committing.
• Not sure? Generate the SRT first. You can always burn it in later; you cannot un-burn.

Doing it, start to finish

1. Open Auto Subtitle Generator and drop in the video. First run pauses to fetch the model.
2. Let it transcribe. Longer videos take longer — your device is doing real work.
3. Read the transcript before you do anything else. Automatic subtitles are good, not perfect (see below).
4. Fix what is wrong. Video Subtitle Editor lets you correct lines and re-sync timings against live playback, which is far easier than editing raw timecodes.
5. Export the SRT, or burn the captions in with your chosen font, colour and position.

Already have an SRT from somewhere else? Add Subtitles to Video skips straight to the burn-in step.

Where automatic subtitles get it wrong

Anyone who tells you this is a solved problem has not used it on real footage. Expect trouble with:

• Proper nouns. Names, brands, places and technical jargon are what speech models miss most, because they are the least predictable words in any sentence.
• Overlapping speakers. Two people talking at once produces one confident, wrong line.
• Background noise. Traffic, a fan, a busy café — accuracy falls off noticeably.
• Accents the model saw little of. Recognition quality is not uniform across the world's English, and it shows.
• Numbers and spelling-out. "Two thousand twenty-six" versus "2026" is a coin flip.

None of this is a reason to skip automatic subtitles. It is a reason to read them before publishing. A five-minute proofread on a machine-made transcript is dramatically faster than typing captions from scratch, which is the actual alternative.

Other languages

The transcript comes out in the language that was spoken. To subtitle the same video for a different audience, Subtitle Translator takes an SRT and translates it, preserving the timings — so you can ship one video with several subtitle tracks rather than several videos.

Worth doing for reach: a video captioned in a viewer's own language holds attention far better than one they are half-following.

A note on long videos

Transcription is compute-heavy, and it is your computer doing it. A two-minute clip is quick. A ninety-minute recorded session on a modest laptop will take real time — that is the honest cost of not sending it to a server farm.

If you only need part of it, cut first: Video Trimmer takes the relevant section, and transcribing four minutes beats transcribing ninety. If the file is also enormous, Video Compressor will shrink it first — see Compress Video in Your Browser for how that trade works.

Do I need an account or an API key?

No. No sign-up, no key, no per-minute billing. The model runs on your hardware, so there is no meter to run.

Is there a length limit?

No artificial one. The practical limit is your device's patience — and trimming first is almost always the better move for anything over half an hour.

Will the subtitles have a watermark?

No. Burned-in captions come out clean, in the font, colour and position you chose.

Is my video really not uploaded?

Really — and it is checkable rather than a promise. Load the tool, disconnect from the internet, and generate subtitles anyway. If the video were going to a server, that would fail.

Can I edit the subtitles after generating them?

Yes, and you should. Video Subtitle Editor syncs your edits against live playback. Fix names and numbers first — those are where the errors cluster.

Next time a video needs captions, skip the upload queue entirely: open Auto Subtitle Generator, let your own machine do the listening, and proofread the names. The rest of the editing bench is at Video Tools.