Add subtitles to a video
This recipe now opens as part of Languages: choose the matching field there. This page stays as a reference for what each step does.
Goal: take a finished video with speech in it and get a new MP4 with subtitles burned in, plus the same subtitles
as .srt and .vtt files you can upload to a video platform. Nothing else about the video changes: no cuts, no music,
no new voice.
Prerequisites
- The video uploaded to the project (Files tab).
- A speech-to-text provider. An admin adds one under Admin, Inference Providers with capability
Transcription(the localwhisperservice that ships with ReelBolt works). Without one the run stops at the first step and says so, instead of returning your video without subtitles. - For another language, a chat AI provider for the translation.
Subtitles in the language that is spoken
Add the Subtitles for your video template (video-subtitles) from New workflow. It has two steps and calls no AI
model:
| # | Label | What it does |
|---|---|---|
| 1 | Listen to the video | Transcribes the speech. Select your video as the source. |
| 2 | Burn in the subtitles | Draws each sentence on screen while it is spoken and writes the subtitle files. |
Say which language is spoken in your brief if detection guesses wrong (the Spoken language control).
Subtitles in another language
Add Subtitles in another language (video-subtitles-translated) instead and put the language you want in your brief
(the Subtitle language control). It adds a step that translates what is said. The timing is always the original's:
the translation only replaces the words, one sentence for one sentence, so a subtitle never appears earlier or later
than the sentence it translates. A sentence the translation missed stays in the original language rather than
disappearing.
What you get
- The video with the subtitles burned in (the run's output).
subtitles.srtandsubtitles.vtt, with the same sentences and timing as the burned subtitles. Cues are at most 42 characters wide, split at word boundaries, held long enough to read. The run keeps both files with the result of its last step. The run page has no download button for them yet; the next section says how to get subtitle files today.
Download the SRT and VTT files today
Until the run page has its own button, there are two ways. Both work from your own account, with nothing to set up.
From the editor. The editor makes subtitle files from the same speech, and you can fix a word before you download.
- Open Editor in the left menu, press New timeline and put your original video on it (the one without subtitles).
- Open the Analyse tab and choose Transcribe speech on the video. It shows what it will cost first and starts only when you confirm.
- Open Subtitles and press Generate captions. Correct any caption in the list.
- Open Deliver. Under Captions, Download SRT and Download WebVTT give you the files straight away (they carry what is saved, so save first). Or choose Separate files only, or Both to burn them in as well, and press Render: the finished film's card then lists the two files with a download button each, and they are added to your project's files.
The editor listens to the video again, so its line breaks can differ a little from the subtitles burned in by the run. It does not translate. More on these panels: The timeline editor.
The run's own files. Use this for translated subtitles, or when you want exactly the lines burned into the video. While you are signed in, your browser can fetch the two files the run wrote:
- On the run page, copy the address of the video's Download button. It contains
/api/v1/projects/<project>/outputs/<result>/download. - In that address, change
outputstostep-results, and replace everything after<result>with/subtitles/srt(or/subtitles/vtt). You get/api/v1/projects/<project>/step-results/<result>/subtitles/srton the same site. - Open the new address in the same browser. The file downloads. If it answers that no subtitles were written, the run's last step did not make subtitles from speech.
If something looks wrong
- The run stops with "found no speech to caption". The transcript is empty: check that the right video is selected, that it has speech, and that the transcription provider works.
- A sentence is missing. Sentences the transcription did not recognise cannot be subtitled; fix the spoken-language setting and run again.
- Subtitles overlap the picture's own text. They move to the top of the frame automatically where the bottom is busy.
- The look. Open the second step to change the style, font, size, colours and position, like any captions.