the text is on screen and you cannot select it

the built-in way to copy text from a video

1. pause the video.
2. hover over the text in the frame — the cursor becomes a text cursor where Live Text has found something.
3. select it and copy as usual, or right-click for the copy option.

this has worked since macOS Ventura, works the same way in 26, and works well. the list of places is short and worth memorising, because everything else here follows from it.

where Live Text actually lives

• Safari — video on a page, paused.
• Photos — anything in your library.
• Preview — a picture you already have as a file.
• QuickTime Player — a local file.
• Quick Look — the preview you get by pressing space on a file in Finder.
• TV — the same player as QuickTime underneath, with an asterisk that gets its own section below.

the recognition under both routes is the one hop points at the screen: text recognition on Mac

that list is not Apple playing favourites. Live Text arrives with three specific pieces of macOS — WebKit, Quick Look and AVKit's video player — and an app gets it by being built on one of them. Safari is WebKit, QuickTime and TV are AVKit, and anything drawing its own video surface gets nothing however good the app is.

Apple, on selecting text in a frame: interact with text in a photo — Apple Support

and the fallback that always works: screenshot with ⇧⌘4, then open the png. Preview and Quick Look both run Live Text on it. two extra steps and a file on your Desktop, but it costs nothing and never fails.

where you cannot copy text from a video

• Chrome, Firefox, Edge, Arc — video in any browser but Safari. this is the biggest one, and the reason the search results here are full of extensions.
• VLC, IINA and other third-party players.
• a screen share in a call: someone is showing you a document you cannot take a line out of.
• a video inside an app — a course player, a desktop client, anything with its own window.
• and beyond video: a locked pdf, a chat screenshot, a scanned page.

a QR code on screen is the same problem in a different hat — camera in your pocket, code on the display: scan a qr code on Mac

the case nobody mentions: protected video

there is one more refusal, and none of the how-to pages mention it: protected playback. frame analysis is switched off for FairPlay content by design: not degraded, not unreliable, off. a rented film, a subscription stream or most of what you open in TV will never hand you a line of text, however clean the frame. if selection works on a YouTube page and does nothing in a streaming app on the same Mac, you have misconfigured nothing.

Apple's engineers, on the property that turns frame analysis on and what it excludes: add Live Text interaction to your app — WWDC22

the engine is already here — try it yourself

the engine itself is not missing from your Mac. it is installed, documented, and you can drive it from a terminal in a minute, no Xcode needed. the Command Line Tools are enough:

1. save any frame as a png — ⇧⌘4 will do it.
2. in ocr.swift, import Vision, make a VNRecognizeTextRequest and hand it that file through a VNImageRequestHandler.
3. run swiftc ocr.swift -o ocr, then ./ocr frame.png — the lines come back as text.

on macOS 26.6.2 that reads a frame cleanly, and asking the request which languages it knows returns thirty. nothing is absent from your Mac. what is absent is a way to point it at a rectangle of screen without writing a program first.

the request that does the reading: VNRecognizeTextRequest — Apple Developer

so: if you live in Safari and Photos, the built-in path is complete. close this tab, nothing is going to beat free and already there. the cost starts where the source is not an Apple app, and it grows with how often you do this.

how hop copies text from a video

press hop's shortcut and drag a box around any part of the screen. the text comes out in a window you can read, fix and copy from, and lands in the clipboard history at the same time. what is underneath does not matter: a paused video in Chrome, a call, a player, a dialog that will not let you copy its own error message.

same reading either way, Vision underneath both. the difference is what you have to know first, and how many times you repeat it:

everything you lift this way stacks up instead of overwriting the last one: find something you copied earlier

• line breaks are kept, so a table or a code snippet comes out readable rather than as one run-on line. this is the difference that matters if what you are lifting is code.
• QR and barcodes are read in the same pass, and a code holding a web address gets an «open link» button. web addresses only: a scanned code is untrusted input, so a phone number or a Wi-Fi password stays plain text you look at first.
• you can drop a picture into the window instead, or paste one with ⌘V, when what you want is already a file.
• Vision runs on this Mac and nothing is uploaded — for somebody else's internal document that is the only acceptable answer.

framing the screen is the one thing here that asks for the screen-recording permission, and only when you frame — a dropped picture asks for nothing: why that permission keeps coming back

when copying text from a video goes wrong

• nothing is selectable anywhere in that app, on any frame. check whether the stream is protected first — on FairPlay content the analysis is off and there is no setting for it.
• the frame is blurry or mid-motion. step one frame at a time if the player allows: recognition reads pixels, and motion blur is unreadable.
• the text is small. go full screen before selecting: at eight pixels tall nothing reads anything reliably.
• light text on a busy background, or heavily stylised. subtitles burnt over footage are the hard case; crop tightly to the line.
• the language is not one of the thirty, or the frame mixes two scripts. accuracy drops sharply on mixed lines.
• you wanted every subtitle in a whole video. wrong tool — that is transcription, and frame by frame will take longer than the video.
• a locked pdf gave up its text and the layout collapsed. columns are a layout, not a property of the pixels — expect to rebuild them.

pulled several frames and the earlier ones seem gone? they are behind the shortcut: clipboard history on Mac

if what you need from that frame is a colour rather than a line of text: get the hex code of a color on screen

frequently asked questions

can macOS copy text from a video?

yes, since Ventura — pause it in Safari, Photos, Preview, QuickTime or Quick Look and Live Text makes the frame's text selectable. not in other browsers, not in third-party players.

why doesn't Live Text work in Chrome?

Live Text comes with WebKit, Quick Look and AVKit's video player, and an app gets it by being built on one of those. Safari is WebKit; Chrome and Firefox draw their own video surface, so there is nothing for macOS to attach to.

why can't i select text in a streaming app or a rented film?

frame analysis is switched off for FairPlay-protected content by design, so most of what you watch in TV or a subscription app will never offer selection. not a setting, not a bug on your side.

how do i copy text from a video without an extension?

screenshot the paused frame with ⇧⌘4 and open the png — Preview runs Live Text on it and you select the text there. costs you a file you then delete.

can i get text out of a pdf that won't let me select it?

yes, by recognising the pixels rather than asking the document. the restriction is a flag inside the pdf, and it governs selection — not what is drawn on your screen.

does text recognition send anything to a server?

in hop, no — Apple's Vision running locally, the same engine you can drive from a terminal yourself. worth checking for anything you point at a work screen: a fair number of these tools recognise in the cloud.

how many languages can it read?

thirty, on macOS 26 — the european set, both chinese scripts, japanese, korean, russian, ukrainian, arabic, turkish, thai and vietnamese among them. mixing two scripts in one line is where accuracy falls off.

will it keep the formatting of a table?

line breaks are kept, so rows stay rows and a code block stays readable. column alignment is a visual arrangement rather than characters, so wide tables need a tidy afterwards.

the tool behind this

done with text recognition

other things people do

• you need the hex. the Mac shows you rgb.color picker
• your Mac has a speed test. it has no window.speed test
• the pdf is 40 mb and the form accepts 10file converter