Look
Captions transcribed on your Mac.
Most demos are watched with the sound off. Lumae transcribes the audio locally with Apple's speech models and turns the timed words into lines a person can actually read at speed, which is a different problem from getting the words right.
- Transcription
- on-device, Apple speech models
- Maximum line
- 40 characters
- Maximum duration
- 3.5 s per line
- Minimum on screen
- 0.8 s
- Break at pauses over
- 0.7 s
- Text width
- up to 82 % of the canvas
Local, because it can be
Transcription runs on your Mac with Apple's own speech models. No audio is uploaded, no account is involved, and there is no per-minute cost, so captioning a long demo is the same as captioning a short one. It is also why Lumae asks for macOS 26: the on-device models it uses set that floor.
Lines built to be read
A transcript is a stream of timed words. A caption is a line somebody has to read while watching something else, and the two are not the same object.
Lumae breaks the stream at sentence ends and at pauses longer than 0.7 seconds, then keeps each line to 40 characters and 3.5 seconds, and holds every line at least 0.8 seconds so nothing flashes past. Lines never overlap: each is cut at the next one's start, and when that cut would leave a line too short to read, its words join the next line instead of being dropped.
Editable, and styled with the video
Every line is editable inline, and captions live on the timeline like anything else: drag them, trim them with handles, split them, snap them to a clip boundary or the playhead.
The style is part of the video's look. Visibility, size as a fraction of the canvas height (so it holds at any export size), position at the top or the bottom, and whether the text sits in a pill.
Rendered with the picture
Captions are rasterised with CoreText into their pill and cached, then composited after the camera move, so a push-in does not drag the text with it. The same renderer draws the preview and the export, so what you read while editing is what lands in the file.
Questions
Does my audio leave the Mac to be transcribed?
No. Transcription uses Apple's on-device speech models. Recording, editing, captioning and rendering all happen locally.
Can I fix a word the transcription got wrong?
Yes, inline on the caption itself. Editing the text does not re-run the transcription or disturb the timing of the surrounding lines.
Can I export a subtitle file?
Not yet. Captions are rendered into the video today. Burned-in captions are what most demo embeds need, and a sidecar file is on the list.
Next
All features
Record your next demo. Annotate your next screenshot.
Lumae is free. Download the disk image, drag it into Applications, and press record or take a screenshot.
See what changed in the latest release