Add captions to a video, free, no watermark
No account needed: an automatic transcript you can search, edit, and export.
Undo is off for this step: the device needed the memory.
Some steps can no longer be undone or redone: the device needed the room.
Your original file can no longer be read. It may have been moved or deleted.
Auto-Captioner options
Drop a video or audio clip here
MP4, MOV, WebM, M4A, MP3, WAV, OGG · English speech · up to 60 s
Captions are made on this device. Your clip is never uploaded, not even its audio.
English speech only · clips up to 60 seconds
Setting up the captioner…
Captioning your clip…
The result is marked unverified and may be blank or wrong.
Have a subtitle file? Burn it in with Burn Subtitles: no speech model, and clips up to 5 minutes.
Use Burn SubtitlesHave a subtitle file? Burn it in with Burn Subtitles: no speech model, and clips up to 5 minutes.
Use Burn SubtitlesTip: Chrome or Edge on a desktop computer covers the widest range of devices for auto-captioning.
Try a different file
MP4, MOV, WebM, M4A, MP3, WAV, OGG (must contain audio)
How to use
- Drop or choose a clip with an audio track (video: MP4, MOV, WebM; audio-only: M4A, MP3, WAV, OGG), or carry one over from another tool on this site without picking it again.
- The first time you use this tool in a browser, it downloads a small speech-recognition model. The page shows its size as it downloads, with a progress bar. After that, it runs from the cached copy with no further download.
- The clip is checked for speech before anything is transcribed. A silent or music-only clip stops here with a plain message, rather than downloading the model for nothing.
- Review the generated captions in the cue editor: search the transcript, edit any line’s text, and preview the clip with the captions overlaid before you commit to anything.
- Choose your output (SRT, VTT, transcript text, a video with the captions burned into the picture, or any combination) and download.
What the promise actually means
No login. No watermark. No install. No limit on how many clips you caption. That’s different from the clip-length limit below (shown on the page before you start), which is about how much of one clip a single run can process, not a quota on how many times you can use the tool. All four of those hold true right up to the moment you export, not just in a preview, and it’s tested that way on every release, not assumed.
Every result also carries one line above the cue editor: “Captions are generated automatically by a speech-recognition model running in your browser and may contain errors. Review before sharing.” It’s not a legal requirement for a plain transcript. It’s what an honest tool says about a model that can be wrong.
Worked example
Input: a 22-second clip of one person speaking clearly, on a desktop browser. Expected output: several timed caption cues, each two lines or fewer, ready to search, edit, and export as SRT, VTT, plain text, or a burned-in video, and a live preview showing the captions over the clip before you download anything.
Input: a 15-second clip of music only, no speech. Expected output: the speech-recognition model never downloads and never runs. The page states plainly that no speech was found, and no caption file (empty or invented) is offered for download. An “attempt it anyway” option exists for edge cases, and its output is clearly labeled as unverified.
Edge case: a clip that isn’t in English may still produce a result, but it’s flagged as an unreliable transcript from an English-only model, reviewed and correctable in the same cue editor, never silently trusted.
Edge case: a clip longer than this run’s clip-length limit is never silently cut down. If it’s only a little over, you’re offered an explicit choice: caption just the part that fits, with an on-screen notice stating exactly how much was covered, or trim the clip first and caption the whole thing. If it’s far over, trimming first is the only sensible next step, and the page says so plainly rather than offering to caption a sliver of a much longer clip.
Limits and formats
| Accepted input | MP4, MOV, WebM, M4A, MP3, WAV, OGG; must contain an audio track |
| Language | English speech only, stated before you start; other languages get a plain warning, not a hard block |
| Outputs | SRT, VTT, transcript text, burned-in video; pick one or several |
| Style controls (burn-in) | Font size (small/medium/large), position (top/bottom), background box on/off, color from a fixed set |
| Maximum clip length | This tool works on a clip up to the length the page shows before you start: the same cap on every device that can run this tool at all, not longer on a fast device or shorter on a slower one. In the rare case where this device had a problem on an earlier attempt, this one run’s limit can be lower for that attempt only, and the page always states the exact number before anything downloads. A longer clip isn’t silently cut down: if it’s close to the limit you’re offered a choice to caption the part that fits or trim first; well past it, trimming first is the offered next step. This is a processing limit for one run, not a limit on how many clips you can caption |
| First-run download | The size the page shows as it downloads (it depends on which build you’re on), downloaded once per browser, then cached |
| Input file size | This tool has a size cap, above which a file is refused with a plain reason instead of a crash. The current limit is shown on the page before you start |
| Limits on this page | Files up to 2 GB; clips up to 60 seconds (15 seconds for one run after a caption run in this browser didn't finish). |
| Cue formatting | Up to 42 characters per line, up to 2 lines per cue |
Privacy
This tool runs almost entirely in your browser. Picking a clip, extracting its audio, checking it for speech, and running the speech-recognition model all happen on your own device using a model file downloaded once from Inkclip‘s file storage. Your clip, its audio, and the resulting transcript are never uploaded anywhere and never seen or kept by Inkclip. What does leave your device: the page and its assets, that one-time model download, an ad request, and visit statistics (the page you open and which tool you use, never your clip), all described in the privacy policy, which also states what a future opt-in server option would send, if one is ever added, and why none is offered today.
Frequently asked questions
Is there a watermark or a login required?
No. There’s no account anywhere on this tool, and nothing added to your downloaded file beyond the captions you asked for.
What happens if my clip has no speech?
The tool checks for speech before it transcribes anything. If none is found, it says so plainly and produces no caption file, never an empty or invented one. An override to attempt it anyway exists, and its result is clearly marked as unverified.
How long can my clip be?
As long as the page’s clip-length cap allows, shown before you start, the same cap on every device that can run this tool at all. Speech recognition is heavier work than the other tools on this site, so it takes longer to process on an older or slower device, but that’s a difference in wait time, not in how much of your clip gets captioned. In the rare case where this device needs a lower limit for one run because of an earlier problem, the page states the exact number that applies before any download starts, so you never wait through a download only to be told afterward that your clip doesn’t fit.
What happens if my clip is longer than that limit?
Never a silent guess. If your clip is only a little over the limit for this run, you get an honest choice: caption just the part that fits, with a plain on-screen notice showing exactly how much was covered, or trim the clip down first and caption the whole thing. If it’s well past the limit, captioning just a small leftover piece usually isn’t worth the download and the wait, so the page offers trimming first as the clear next step instead. If you do choose the shortened option and later burn the captions into a video, the full clip is kept. The captions simply stop at the point that was covered, and the page says so before you start.
What if my device can’t run this at all?
A small share of devices can’t run the speech-recognition model. When that happens, the page says so in one plain sentence and offers burn subtitles instead. If you already have or can get a subtitle file for the clip from elsewhere, that tool renders it into the video using only your browser’s built-in video support, no model required.
Can I edit the captions before downloading?
Yes. Every cue’s text is editable inline, the whole transcript is searchable, and you can preview the clip with captions overlaid before committing to any export.