Transcripts and subtitles

Yes. Every new upload with sound is transcribed automatically after it's checked. The card shows Transcribing…, then Transcript. For older recordings (or if it didn't run), use ⋯ → Transcribe. Transcription usually takes about 1–3 minutes per hour of speech once it starts (a 30-minute recording took about 1½ minutes in our tests). Only the sound is used: it's compressed, cut at pauses into pieces of about 10 minutes and transcribed at the same time by our speech service (Whisper, on Groq), then the pieces are put back together with exact timings. If that service is busy or down, your recording is transcribed on our own servers instead; nothing changes for you except that it may take a little longer.

When you use an edited or joined video again (Edit more, or as a clip in Join), it gets its words straight away from the recordings it was made from: each word is moved to where it plays in the edited video (after an intro, without the parts that were cut, faster or slower with Speed). Nothing is transcribed again and it never counts toward your video minutes. Edited videos are never transcribed or sent to the AI on their own (most are never edited again). When you open one in Finish my video, a card asks whether you want AI help and shows how long each step takes ("Write down the words", then "AI suggestions"). Press Start, or Continue without AI. If the words were copied from your clips, only the AI step runs. You can also press Get AI help later on the same page.

A problem in October 2026 made some transcriptions fail. Those recordings are transcribed again automatically, a few at a time; you can also use ⋯ → Try transcribing again.

⋯ → View transcript opens the player with the text. Click any line to jump there; the line being spoken is highlighted. You can search, copy all text, or Download as Text (.txt), Subtitles (.srt) (YouTube, Facebook, most editors) or Web subtitles (.vtt) (web players, LinkedIn). Subtitle lines are at most two lines of 42 characters and 6 seconds each.

The language is detected automatically. Around 100 languages can be transcribed; the list in Transcribe as… shows the common ones (including English, French, Spanish, Portuguese, German, Yoruba, Hausa, Swahili, Arabic, Hindi, Chinese and more).

Open ⋯ → View transcript and, at the bottom, choose Wrong language? → Transcribe as… and pick the right language. If the recording was already transcribed in that language before, Davnai switches back to it instantly. Otherwise it's transcribed again in the new language (the card shows Transcribing…). The newest transcript is the one shown and used for AI edits. This never counts toward your video minutes.

Transcribing the same recording again with the same language gives exactly the same words, so it would only waste time. The only reason to redo it is a wrong language, which Transcribe as… handles. If transcription failed, the menu shows Try transcribing again.

The recording has music, sound effects or silence but no talking (for example an ad or B-roll clip). Every tool still works on it (Join, Resize, music, logo, call to action); AI edit suggestions are skipped because they work from what people say.

The recording has no sound track, or no speech was found. Check that the original has audio.

You can fix a word that was heard wrong (owner, 25 October: "Flatton" should be "Fladon"). In the video editor's Words (and the AI Director's Words tab), double-click the word, type the right spelling, and choose This one or All in this video. Find and fix a word fixes it everywhere at once. Captions, the Words, posts, chapters and downloaded subtitles all use the fixed spelling; your voice isn't changed. Undo brings back the old word. Tick Always spell it this way and it goes in Brand Kit → Your words, so new recordings are written right first time. Rewriting whole sentences isn't possible (the words must match what's said). Translated captions can be fixed too: click the word on the paused video with Captions open.