Rev's job is an accuracy pass: a human-checked transcript, then whatever you do with the words. Versely splits that same sequence into two named tools rather than one styled caption render. transcribe_audio (the get-a-video-transcript task) returns editable text. add-captions-to-video burns a caption track. Style is a later decision. The point of this page is the gap between those two calls — the read-back — not a preset picker.
What each step actually does, and why running the burn first wastes a caption pass on words you have not checked, is below.
Plain text first, from transcribe_audio
get-a-video-transcript runs Cartesia's ink-whisper model on an audio_url and hands back a document. transcribe-audio-to-text is the same call from the agent. Language can be set or left to detection.
That document is for reading. It is not burned in. It is not a caption track. Accuracy depends on audio clarity — background noise, overlapping speakers and heavy accents reduce it, which is why the next step exists.
The quality pass is you, on the document
Names, numbers and homophones fail first. Accents do not have an honest WER you can quote from a vendor card. Read the transcript against the take. Fix the line on the page before any captioner typesets it.
There is no separate human-captioner queue inside Versely. The QC is the edited document. If a line is wrong on the page, it will be wrong on the picture.
Then burn, on the keeper
add-captions-to-video (add_veed_captions) times speech to the picture and skins the lines. transcribe-and-caption-my-video is that burn as one instruction — use it after the document has been read, not instead of the document.
How-to for feeding a corrected script back as timed lines is add_timestamped_captions, which takes your wording and your clocks rather than re-listening. That is the path when the ASR line and the true line have diverged.
Do not shop a look on an unread transcript
A caption style does not fix a wrong word. Pick the family after the text is stable. Burning first and restyling later recaptions the whole file; burned-in captions are pixels.
How it works
1. Transcribe to a document
get-a-video-transcript / transcribe_audio. Audio or a video's soundtrack. Editable text out.
2. Read it back
Names, numbers, homophones, crosstalk. Fix the line on the page. This is the accuracy pass.
3. Burn the keeper
add-captions-to-video if the ASR lines held. add_timestamped_captions if you are typesetting the corrected wording yourself.
4. Choose a look only after the words are stable
A preset change on a burned file is a new render. Do not use style as a substitute for the read-back.
Where this lives in Versely
Who this fits
- Interviews where proper nouns have to be right on screen
- Training and course clips that cannot ship a hallucinated number
- A caption burn you want to QA as text before it becomes pixels
- Feeding a corrected transcript into timed overlays when ASR and the true line have split
Frequently asked questions
How does Versely compare to Rev?+
Versely runs an editable transcript first (transcribe_audio / get-a-video-transcript), then burns captions on the keeper (add-captions-to-video). The quality pass is reading that document — names, numbers, homophones — before any style is chosen. There is no separate human-captioner queue; the corrected page is the QC.
Is the transcript already a caption track?+
No. transcribe_audio returns plain text. Burning is a second tool. Mixing those two jobs is how you typeset a line you have not checked.
What if the ASR line is wrong?+
Fix it on the document. If the true wording no longer matches what the captioner would hear, stamp the corrected lines with add_timestamped_captions instead of re-running speech-to-text.
Which engine writes the document?+
Cartesia's ink-whisper model, on the audio you point it at. Clarity of the source still governs accuracy.
Other alternatives on Versely
Further reading
Try it inside Versely
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync, in your browser or on your phone.
Free on the web and on iPhone, one account.
Reviewed September 4, 2026. Facts about Rev on this page are general, publicly known positioning, not pricing or feature claims — see /alternatives for how this page set is scoped. Versely capability links above are pulled from the same live data the rest of versely.studio uses, so they move when the product does.