Auto captions, no watermark

how it works

How to add captions to a video

Drop your clip above, let the speech-recognition model transcribe it right in your browser, fix any words it misheard, pick a lower-third style, and export with the captions burned in. Free, no upload, no account, no watermark.

1. Drop the video — transcription runs locally in your browser.
2. Edit the caption lines; timing stays attached to each line.
3. Choose one of the lower-third styles.
4. Burn in and download the captioned video.

what to check

Where speech recognition slips

Whisper is strong on ordinary speech and weak in a few predictable places. Reading the transcript once with these in mind catches nearly every error, and each line stays clickable so a correction takes seconds.

terminology

Captions, subtitles and open captions

The three terms describe different jobs. Captions carry the audio in the same language for viewers who cannot hear it, sometimes including sound cues like [door closes]. Subtitles translate dialogue for viewers who do not speak the language and assume the viewer can hear. Open captions are burned into the pixels rather than delivered as a separate file, which is what this tool produces.

Burned-in captions cannot be turned off, which suits social feeds where video plays muted by default and caption settings vary by viewer. The trade-off is that they cannot be translated later, resized for a different aspect ratio, or read by search engines the way a caption file can. For a video you will re-cut or localise, keep the caption text as part of the project rather than baked into a finished export.

readability

Captions people can actually read

A caption has to be read in the time it is on screen, by someone holding a phone at arm’s length, often in daylight. That makes readability a timing problem as much as a design one. Comfortable reading runs at roughly 15 to 20 characters per second, so a 40-character line needs two to three seconds on screen. A line that flashes past faster than that is decoration, not communication.

Two lines is the working maximum, at about 32 to 42 characters each. A third line starts covering the picture, and on a vertical video it pushes into the region where the platform draws its own interface. Break lines at natural phrase boundaries instead of filling to the character limit. “We cut the render time / from forty minutes to four” reads cleanly; a break after “from” forces the viewer to hold an incomplete thought across a line change.

Contrast decides whether any of that matters. White text on light footage is unreadable outdoors regardless of size, so every usable caption style carries either a solid backing bar or a heavy shadow instead of relying on text colour alone. Size scales with the smallest screen you expect. Text that looks generous on a laptop preview is often marginal on a phone, and phones are where most of this footage gets watched.

placement

Where the platform UI will cover your captions

Vertical feeds draw their own controls over your video, and they draw them in the same places every time: a caption sitting at the bottom of the frame lands under the account name, the description and the sound label, while the right edge carries the like, comment and share column. Anything you place there is competing with interface you do not control.

why bother

What captions are actually for

The usual argument is that feed video plays muted, and that is true enough to matter. But captions do three other things that get less attention. They make the video usable by people who are deaf or hard of hearing, which is a requirement, not a nicety. The WCAG guidelines list captions for prerecorded audio as a Level A criterion, the baseline tier, and public-sector procurement in several markets checks for it.

They also help people who can hear perfectly well. Anyone watching in a second language, in a noisy office, on a bad connection where audio drops before video, or following a technical term they have never seen written down. A spoken product name is a guess until it appears spelled out. For software demos in particular, captions are often the only place a viewer learns how a command or a flag is actually written.

The third is retention. Captions keep a viewer anchored during the pause between spoken sentences, which is exactly where people scroll away. The format spread from accessibility tooling into mainstream editing because it helped the median viewer, not only the ones who needed it. Burned-in captions of the kind this tool produces cover all three cases at once, with the trade-off that they cannot be switched off or translated later. For that, keep the caption text in the project and re-render instead of baking it into a finished file.

The common mistake is treating captions as a transcript. A transcript records every word, including the false starts, the repeated phrases and the “so, basically” at the head of a sentence. Captions are read, not heard, and reading is less forgiving than listening. Filler that disappears in speech becomes clutter on screen, and it eats the seconds a viewer needs for the words that matter. Trimming a spoken line to its meaning is the same edit a subtitler would make, and it is the difference between captions that get read and captions that get ignored. The exception is anything quoted or legally significant, where the exact words are the point.

Speaker labels are the other judgement call. On a single-narrator screen recording they are noise. On an interview or a podcast clip, where two voices alternate and the speaker is often off camera, a short label at the start of each turn is the difference between a viewer following the conversation and giving up on it. Keep them to a first name or a role, and only where the change of speaker is not obvious from the picture. A caption announcing who is talking while that person is plainly on screen spends characters the viewer needs elsewhere. The same rule applies to sound cues: [laughter] earns its place when it explains a reaction the viewer can see but not hear, and clutters the frame when it does not.