Auto Captions: Generated Subtitles Explained

There are two ways to get them, and they fail differently.

Generated from the audio

Speech recognition listens to the finished track and writes what it hears. It works for any video, including one you filmed. It also makes the mistakes recognition always makes: names, acronyms, jargon and any word spoken over music.

Every faceless niche has a vocabulary that trips it. A history channel gets place names wrong, a finance channel gets tickers wrong. Those words sit on screen, large, for everyone to read.

Generated from the script

When the audio came from text to speech, the words are already known. The model returns the timing of each one, and the captions are built from that rather than guessed. No recognition step, no misheard words, and the highlight lands on the spoken syllable.

This is the better path whenever it is available, and it is why script-first pipelines produce cleaner captions than recording first and transcribing afterwards.

Checking them before you post

Read every proper noun and number.
Watch the last 5 seconds. Drift accumulates, so the end is where sync breaks first.
Check the placement against safe zones.
Confirm the style survives a busy background. See subtitle burn-in.

Auto captions are a first draft that is right most of the time. The ten seconds it takes to scan them is the difference between a professional video and one with a misspelled name in 90-point type.

MakeViral builds captions from the voice model's character timings. See the AI voiceover video generator.

FAQ

Questions about auto captions

Are auto captions accurate enough to post without checking?

For ordinary speech, usually. For names, numbers, acronyms and niche vocabulary, no. Those are exactly the words that carry the meaning of the video, and a mistake in one of them sits on screen in large type for the whole clip.

Why do my captions drift out of sync near the end?

Because the timing was estimated from an average reading speed rather than measured from the audio. Small errors accumulate, so the first few seconds look fine and the last few are visibly late. Captions built from the speech model's own word timings do not drift.

Should I use the platform's caption tool or burn captions into the video?

Burn them in. Platform captions can be switched off, get restyled by the player, and disappear when a video is downloaded and reposted. Burned-in text is visible to every viewer in every context, which is what a voiceover-driven video needs.

See it in action

Make one, do not just read about it

Generator

TikTok Voice Generator for AI Voiceover Videos

The narrator-over-gameplay format: a written script becomes a voiced, captioned vertical video with a background clip, in one pass.

Open it

Keep going

Related terms

Glossary

Subtitle burn-in

Subtitle burn-in means rendering captions permanently into the video frames, so the words are part of the picture and cannot be switched off, moved or restyled by the player.

Text to speech (TTS)

Text to speech, or TTS, is the technology that converts written text into spoken audio, producing a narration track from a script without anyone recording it.

Auto captions