Reference
Automatic transcription on your Mac
Last updated November 10, 2026 · Read as Markdown
To transcribe video on your Mac without uploading it, add the folder or drive with your footage to Grain Photo. Grain transcribes every recording on your Mac, using Apple's speech processing in macOS 26, so every recording on every drive gets a timed transcript you can read, search, correct and add clips from.
How it works
Starts by itself
- Begins once a recording is indexed
- Pauses and resumes around your work
- No button, no queue, nothing to retry
Stays on your Mac
- Apple's speech processing, on your Mac
- No upload and no transcription account
- Files stay where they are
Timed to the picture
- Search results play from the right line
- Clips pick up the right words
- Exported captions line up with the video
The transcript is kept in your Grain library. The original recording isn't changed, and no sidecar file is written next to it.
How transcription starts
You don't start it. Click a folder or a drive under Sources, and Grain catalogs the files in place. When a recording has been indexed, transcription begins in the background, and you can browse, review and search while it continues.
There's no Transcribe button. New recordings you add to a source are transcribed the same way. See add and navigate sources if you haven't added your drives yet.
What it needs
- macOS 26 on an Apple silicon Mac.
- Apple's speech resources. Grain uses Apple's speech processing, so it needs these on the Mac.
Read a transcript in Moments
Open a recording and Moments shows its transcript in order, beside the player. Click a passage to play the recording from there. Previous and next step through every moment in the recording.
All
- The transcript and your clips together
- Each clip in the context of what was said
Clips
- Only the clips you've added
- From this recording
Show All
- The full reader
- Read a long interview top to bottom before you watch any of it
When you open a recording from a search, the passages that match are highlighted, with everything around them still visible. To search inside one recording, use Search Moments: the list narrows to the moments that contain your words. To look beyond a passage, scrub the video timeline.
Reading first is often faster than watching. You can skim an hour-long interview in a few minutes and know which ten minutes to listen to.
Correct a misheard word
Transcripts are good, not perfect. Names, places, jargon and mumbled lines are the usual misses. Double-click a moment's words to correct them, and your correction is used everywhere the words are:
- Search finds the corrected word, so the next search for a client's name lands on it.
- Captions in Grain use the correction.
- Exports use it too, so the SRT that comes with an exported clip is right.
Search what was said
Once a recording is transcribed, what was said becomes part of your library. You can search footage by what was said: exact words or a keyword. Dialogue search finds the passages where they're said and puts the closest matches first.
Search one source, a collection or All Media. A result plays from the matching moment, with a little context before it, and pauses at the end of the passage.
Transcripts and clips
A clip is a range linked to the original recording. If you add a clip before transcription has reached that recording, for instance with I and O while indexing is still running, the clip picks up its words when transcription finishes. See add, trim and export clips for the full clip workflow.
Shortcuts
| Keys | What they do |
|---|---|
| Space | Play and pause |
| , . | Step back or forward one frame |
| I O | Mark the start and end of a range |
| ⌘Return | Add the marked range as a clip |
| Esc | Cancel a range you're making or editing. Otherwise, leave the video |
Every shortcut in Grain is on one page: keyboard shortcuts.
Privacy: what stays on your Mac
Transcription runs on your Mac with Apple's speech processing. Your footage isn't uploaded to transcribe it, and the transcripts are kept in your library, not in a cloud account.
If you connect an agent such as Claude or ChatGPT, it can search spoken dialogue across a source and read a recording's timed transcript. Whatever the agent reads, including transcripts, previews and metadata, goes to that assistant's provider under its policies. Leave the agent disconnected for work you can't share, or keep its requests to sources you're comfortable with. More in how Grain handles your files.
The agent proposes; you make the corrections in Grain.