Turn a video or an audio file into captions on your Mac - transcribe, correct the wording, time it properly, style it, and export the format the next tool expects. Nothing is uploaded at any stage.
Start on Universal Lite, which ships with the app and handles everyday clips. Move to Universal Compact for longer talking-head video, and to Universal Plus or Max when the audio is genuinely hard - a heavy accent, several people talking, music under the dialogue. On macOS 26 the Instant engine uses the system recogniser and starts with no download at all, in ten languages.
Do I have to download a model before the first subtitle?
No. A speech model is bundled inside the app, which is the whole point of Universal Lite existing. You can subtitle something the minute it opens.
What files can I drop in?
Video as MP4, MOV or M4V, and plain audio as MP3, M4A or WAV. Drag it anywhere in the window. Audio-only files still produce timed captions, which is what you want if the video is being cut elsewhere.
What is characters per second, and what number should I aim for?
It is how fast a line asks the viewer to read. Broadcast subtitlers use it as the standard sanity check, because a caption can be perfectly accurate and still flash past unread. The editor shows it per line so you can see which ones need splitting or shortening rather than guessing.
How do I fix timings that drift?
Click a line to jump to that moment in the video, then snap the timing to the playhead. Merge two lines that were split mid-sentence, or split one that runs too long. The waveform timeline is there to make the boundaries visible rather than approximate.
Should I export a file or burn the captions in?
Burn them in for TikTok, Reels and Shorts, which ignore subtitle files. Export a file for YouTube, a client handover or an editing timeline - SRT is read by Premiere Pro, Final Cut Pro, DaVinci Resolve and VLC.
What is the difference between SRT, VTT and ASS?
SRT is the plain, universal one: text and timings, no styling. VTT is the web equivalent, for HTML5 players. ASS carries your fonts, colours, position and karaoke tags, so it is the one to use when the look matters and the player supports it. TXT is a transcript with no timecodes at all.
How does translation work?
It uses the translation built into macOS, which runs on device and costs nothing per word. Translate a finished transcript and every line stays editable afterwards - worth doing, because machine translation of spoken language usually needs a pass by someone who speaks it.
Can I make the captions look like a specific channel style?
Start from one of the six presets - Classic, Bold Pop, Karaoke, Neon, Minimal or Cinema - then change font, size, colour, shadow, background and position. The preview updates on the video as it plays, so you are judging it in motion rather than on a still.
What does the free version cover?
Transcription, the full editor, every style, and three subtitle file exports. Burned-in video carries a small watermark. Pro removes the watermark, makes exports unlimited, and adds the larger engines for the hardest audio.
Get AI Subtitle Generator: SRT
Generate, edit, style and export subtitles for video on Mac - SRT, VTT, ASS or burned in, entirely on device.
The workflow for something going to TikTok, Reels or Shorts, where the file has to carry the captions itself.
Drag the MP4 into the window and let Universal Lite run.
Read the transcript once for names, jargon and numbers - that is where recognition slips, and it is what viewers notice.
Check the reading speed column and split any line that is too fast.
Pick a preset, then adjust size and position so nothing collides with the platform UI at the bottom of the frame.
Turn on karaoke if the channel uses it, then render a burned-in MP4.
Keep captions clear of the lower third of the frame. Every short-form platform puts its own interface there, and a caption behind a username is a caption nobody reads.
Here the deliverable is the file, and it has to drop into their timeline without argument.
Import the same cut they have, so the timings line up with their sequence.
Transcribe with Universal Compact or better if the recording is long.
Correct wording and timings, paying attention to the first and last lines, which are the ones that get noticed.
Export SRT rather than ASS unless they asked for styling.
Name the file to match the video file exactly - most editors auto-match on filename.
If the cut changes after you deliver, the SRT is wrong from the first edit point onwards. Re-run against the new cut rather than nudging timings by hand.
The case where an upload-based service is not an option at all.
Disconnect from the network if you want to prove the point to yourself or to a client.
Import the footage and transcribe - everything runs on the Neural Engine locally.
Edit and style as normal.
Export the file or render the video.
Keep the project in the library; it stays on the machine with its transcript and styling.
This is the practical argument for on-device work. It is not that a cloud service is careless - it is that "the footage never left this Mac" is a statement you can make without qualifying it.
You’re browsing inside Instagram, and its built-in browser isn’t allowed to hand
links over to the App Store. Open this page in Safari and the link will work.
Tap •••the three-dot menu
at the top right, then choose Open in external browser.