Videos

Add captions to a video

Most tools transcribe the audio to make captions, which means the captions inherit whatever the speech recogniser heard. Here they come from the other direction: the narration text you wrote is the source, the speech is synthesised from it, and the captions are generated from the same string.

Create a video with OneCraftWhere captions come from, how they are timed and drawn, and what they cannot do

The delay, speed and captions row of the scene voiceover dialog, with the captions switch turned on. — Captions are on by default and switch off per scene, sitting beside the delay and the speed for that scene's narration.

Written text in, captions out

Because the narration text already exists as a string, there is no transcription step and therefore no transcription error. A product name spelled correctly in the narration is spelled correctly in the caption. Change the narration and the captions change with it automatically, since they are derived live rather than stored as a separate asset that can drift out of date.

They type on, slightly ahead of the voice

A cue does not appear all at once. Letters reveal progressively across the cue's window, running a little ahead of the spoken line so the word is on screen just before it is heard rather than just after. It is a small thing that makes a muted watch feel deliberate rather than like a subtitle track bolted on.

One fixed pill, never theme flipped

The caption sits in a dark translucent pill with a blur behind it, centred at the bottom, capped at 72 percent of the frame width, in white text at 600 weight. That pill is deliberately not tied to your palette. Captions overlay arbitrary scenes, including full bleed photographs and light backgrounds, and a caption that recoloured with the theme would eventually land white on white.

On by default, and switchable per scene

Captions default to on for every scene and are toggled individually in that scene's voice dialog, alongside the narration delay and playback speed. Turning them off for one scene is useful where the scene's own on screen copy already says the line, and a caption underneath just repeats it in smaller type.

How it works, in three steps

Step 1
Write the narration properly
The caption is the narration, so spelling, product names and punctuation in the narration field are what appear on screen.
Step 2
Leave them on for feed placements
Facebook and LinkedIn autoplay muted. Reels and Shorts play with sound. Captions cost nothing on the sound on surfaces and carry the video on the muted ones.
Step 3
Turn one off where it duplicates
On a scene whose whole design is one big line of type, the caption repeats it. That is the case for using the per scene toggle.

The full walkthrough with screenshots is in the guide Add voiceover, captions and music to a video.

Limits worth knowing

Captions are burned into the picture. There is no SRT or VTT file to upload alongside, so a platform's own selectable captions are a separate job.
Position is fixed per format: 44 pixels up from the bottom in landscape, 120 in vertical, 56 in square. It cannot be moved.
Type size is fixed at 26 pixels in landscape and 30 in vertical, and the font is the video's own family.
There is no word by word highlight styling, no speaker labels and no custom caption colour.

Questions people ask

Do captions appear if there is no voiceover?

No. They are generated from the narration text, so a video built with the voiceover switch off has no narration and therefore no captions. The on screen copy is doing that job instead.

Will they clash with the platform's interface?

In the vertical frame they can. The pill sits 120 pixels up from the bottom, which is above a phone's own navigation but inside the bottom exclusion zone Meta publishes for Reels, so a caption can land under the caption and profile block there.

Can I caption in a different language from the voice?

Not independently. The caption is the narration string, so it is always in whatever language that scene's narration is written in. You can set language per scene, which is a different thing.

Are they accurate to the audio?

By construction, yes, because the audio is synthesised from the same text. The one thing to watch is playback speed: at 1.25 times the words arrive faster than the typing effect assumes on a very long line.

Make your own video

The button opens the generator with this use case already described. Change the wording to match yours, generate, then edit anything you like.

Create a video with OneCraft

Slideshow video with narration

The reason most photo slideshows are unwatchable is that every picture gets the same three seconds and one voiceover runs across the lot. The interesting shot flies past, the filler lingers, and the narration is talking about image two while image five is on screen.

AI video script generator

A video script that reads well on a page usually plays badly. The reason is that the page has no timing. Here the script is produced scene by scene, and every line arrives attached to a duration, an on-screen headline and a layout that has already been chosen.

Background music for video

Music here is one track laid across the whole video, looping to fill it and automatically dropping out of the way whenever anybody speaks. That is a deliberately simple model, and knowing where its edges are saves you fighting it.

More finished work of this kind is on the video examples hub.