How to burn captions into a video, and drop the background

Captions from a script beat captions from audio

Automatic captioning normally works by transcribing what it hears, which means every proper noun, product name and technical term is a guess. If you read from a script, the exact words are already known, so the only thing left to work out is timing.

That is why script-derived captions come out clean: no misheard names, no invented punctuation, no pass to correct afterwards.

Burned in or a separate file?

• Burned in means the words are part of the picture. They always show, on any platform, with no upload step. They cannot be turned off or translated.
• A sidecar file keeps captions separate, so viewers can switch them off and platforms can translate them. It needs somewhere that accepts the file.

For social video, burned in wins, because autoplay is silent and most feeds will not read a sidecar. For anything hosted where accessibility matters, provide both.

Removing the background without a green screen

Modern on-device segmentation separates you from the room without any screen at all. It is not quite as clean as a properly lit green screen, and it does not need two metres of fabric, a stand and somewhere to store them.

1. Put a light on your face, not just on the room. Segmentation follows edges, and edges need contrast.
2. Avoid loose hair against a busy background, which is where every matte algorithm struggles.
3. Sit further from whatever is behind you. Depth helps as much as light.
4. Check a moving take, not a still frame. Artefacts show up in motion.

If the matte flickers at the shoulders, the usual cause is a background the same brightness as your top. Change the top, not the setting.

Both of these live in the same recording flow as the prompter, covered in the Mac teleprompter guide.