Automatic subtitles for video

Subtitles are the most common and at the same time the "cheapest" way (in compute) to finish a ready video: here the network draws nothing but listens to the sound and turns speech into text. So it's a convenient entry point into video processing — fast, cheap and almost always useful.

How it works

At the core is speech recognition (more on this in the guide on speech synthesis and recognition): the network listens to the track, transcribes the words and sets the timings — which phrase at which second. The output is either a subtitle file (.srt/.vtt) or subtitles already "burned" into the frame.

Since the model works with sound, not the picture, subtitles are cheap to compute — this isn't "heavy" video generation.

Upload a clip — get ready subtitles with timings. Speech recognition is cheaper than video generation; the first transcriptions are available after signing up and onboarding.

Why subtitles at all

• Watching with no sound. Most of the social feed is watched without sound — without subtitles you lose those viewers.
• Accessibility. For people with hearing impairments.
• Retention and SEO. Text in the frame holds attention, and a transcript helps platforms understand what the clip is about.
• A basis for translation. Ready text is the first step toward dubbing into another language.

To make it tidy

• Check names and terms. Recognition errs on rare words, brand names, names — they're worth proofreading.
• Watch the line length. Two lines, no more; cut a long phrase by meaning.
• Punctuation and case. Modern models set them themselves, but a quick look-over doesn't hurt.
• Style for the platform. For social — large "burned-in" subtitles; for YouTube — a separate file so the viewer can turn it off.

What's next

Subtitles are the text of your speech. The next step is logical: translate that text and voice the video in another language — that is, make a dub.

In the Twelver chat you can upload a clip right into the conversation and get subtitles — no separate apps. Transcription is available after signing up and onboarding.