How to Extract Subtitles from TikTok, Reels & Shorts (Burned-In Captions) on Mac

TL;DR: The captions on a downloaded TikTok, Instagram Reel, YouTube Short, or Douyin clip are burned into the picture, so there is no subtitle track to copy or download — the only way to get the text back is to read it off the frames with OCR. GeekLink does this locally on your Mac: set the caption area once, batch-process a whole folder of clips, export a timed SRT per video, and translate into 40+ languages for reposting. Free to extract; macOS 13+ on Apple Silicon.

If you have ever tried to pull the captions off a short-form video and found there was nothing to pull, this is why — and here is the workflow that actually gets the text out.

Why can't you just copy the captions off a TikTok?

Because on a downloaded short-form clip, the captions are not text — they are pixels. When a creator adds on-screen captions in TikTok, CapCut, Reels, or Douyin and exports the video, that text is painted directly into every frame, the same as a logo or a face. Your computer sees a picture, not characters, so there is nothing to select, copy, or export.

This is why every "download the subtitles" tool comes up empty on short-form video: they look for a subtitle track inside the file, and short-form clips almost never have one. The captions are real, visible, and completely locked into the image. Getting them back out is a video-analysis job, not a copy-the-track job.

Are TikTok, Reels, and Shorts captions a subtitle track or part of the video?

Almost always part of the video. Short-form captions are burned-in (also called hardcoded or open captions): baked into the pixels, always visible, no on/off switch. This is different from the toggleable subtitle tracks you get on Netflix or in an MKV file, which are separate text you can extract in one command.

There is one narrow exception: if you still have the original export and the creator used a real caption track, you could pull that track directly. But the moment a clip has been downloaded, screen-recorded, or reposted — which is the normal case for anything you did not make yourself — the captions are burned in and OCR is the only way. If you are unsure which kind you have, our embedded vs burned-in guide shows how to tell in ten seconds.

How do you extract burned-in captions from short-form video on Mac?

You need a tool that reads the text off the frames and reconstructs the timing. GeekLink is a native Mac app that scans the video, recognizes the on-screen captions, works out when each line appears and disappears, and exports a time-synced SRT — all locally, with no upload. The steps:

1. Import the clip (or a whole folder of them). GeekLink takes the downloaded MP4/MOV directly.
2. Draw a box around the caption area once. Short-form captions usually sit in a consistent band (center or lower third), so one box tells the engine where to look and what background to ignore.
3. Run it. You get an editable SRT with timings, plus per-line confidence so you can see which lines to double-check.

The whole point is that the text comes back editable. Once it is an SRT, you can clean it up, translate it, re-time it, feed it to a language-learning tool, or burn a new styled version back into the video. It opens in any subtitle editor, so nothing locks you in.

How do you handle animated captions and vertical video?

Short-form captions are the hardest kind of burned-in text, and it is worth being honest about that. The tricky cases:

• Word-by-word "karaoke" captions that pop one word at a time — great for retention, awkward for OCR, because the "line" is constantly changing. GeekLink reconstructs these into readable cues rather than one cue per word.
• Emoji and stickers mixed into the caption line — filtered out so you keep the words, not the picture characters.
• Captions that move around the frame or change color per clip — this is where drawing the caption region and, if needed, filtering by text color keeps clutter and background text out.
• Vertical 9:16 video — handled natively; the caption box just follows the tall frame.
• Low contrast over busy footage — the hardest case for any OCR. Accuracy dips here, so treat the output as a clean draft to proofread rather than a finished file.

The realistic goal is a clean draft you tidy up in a minute, not a perfect file straight out of the tool — and GeekLink flags the lines it was unsure about so you review those instead of re-reading everything. The full review workflow is in finding the lines AI got wrong.

How do you translate short-form captions for reposting to other platforms?

This is where extraction pays off. A huge amount of short-form content is repurposed across languages and platforms — Douyin clips reposted to TikTok, Reels, and Shorts for a global audience, and vice versa. Once the captions are an SRT, GeekLink can translate them into 40+ languages using the model you choose (Claude 3.5 Haiku, GPT-4o, GPT-4o mini, or DeepSeek), so a Chinese Douyin clip becomes ready-to-post English, Spanish, or Portuguese captions.

The end-to-end path for a repurposer: extract the burned-in captions, translate the SRT, then either hand the SRT to your editor or burn the translated captions back into the video. It keeps the original creator's words and timing while swapping the language — far faster than re-typing captions by hand for every market.

Can you batch-process a whole library of clips?

Yes, and for short-form volume this is the part that matters. GeekLink batch-processes 50+ videos in one run: import the folder, set the caption region once on the first clip, and it works through the rest unattended, exporting one SRT per video. For a creator sitting on hundreds of clips, that is the difference between a viable localization workflow and giving up.

Budget-wise, extraction runs on the free tier for trying it out; full batch use plus translation is $99/year (about $8.25/month) or $169 one-time lifetime, both including 1M AI translation credits. Requirements: macOS 13.0+ on Apple Silicon (M1–M4).

FAQ

Disclosure: GeekLink is our own Mac app; the burned-in caption OCR, region and color filtering, confidence flagging, batch processing, and translation described here are GeekLink features.