AI caption generator for video and audio

What the AI actually does

Captions are generated by ElevenLabs Scribe v2, a speech recognition model trained for transcription rather than for conversation. It returns a timestamp for every individual word, not just for each caption block, which is what makes the timings line up with speech instead of drifting a second or two out over a long recording.

That word-level output is also why re-timing is cheap. When you split or merge a caption in the editor, the new block inherits real timings from the words inside it — nothing is interpolated or guessed.

Where automatic captions usually go wrong

The failure modes worth knowing about are accents, overlapping speech, background music and code-switching — a speaker moving between two languages mid-sentence. Scribe v2 handles all four noticeably better than the free auto-caption features built into video platforms, but none of them are solved problems.

This is why the editor exists and why captions are never exported automatically. Read the transcript before you publish it, especially for proper nouns, brand names and numbers, which are where recognition errors concentrate and where they are most visible to a viewer.

Captions in one language, or translated into another

Transcription keeps the spoken language. Translation takes the same audio and produces captions in any of the 76 supported languages, so a Hindi interview can become English captions, or an English webinar can become Spanish ones, in a single pass.

Translation runs on the transcript rather than on the audio, so timings are preserved exactly — the translated caption appears at the moment the original words were spoken.

What it costs

Credits cost $10 for 100, one credit per minute of audio or video, rounded up to the next whole minute per file. A 12-minute podcast episode costs 12 credits. There is no subscription, credits do not expire while the account stays active, and there is no free tier — every account starts at zero.