Stop screenshotting TikTok overlays. Read them instead
A competitive intel lead still keeps a folder called tiktok-overlays. Pause, screenshot, rename with the date, dump into Slack. By the end of the week the folder is a comic of prices and codes that never became rows.
Stop doing that. The overlay is already text. You are treating a video editor's typography like a photograph because no one on the team wanted to sit through another "we should OCR this" ticket.
The folder is a museum of offers you cannot query
TikTok is where a lot of brands now publish the thing they used to put on a landing page: the drop name, the percentage, the "today only," the location of the pop-up. It is typeset on the video. The native caption is a joke or a hashtag. The audio is a sound they licensed.
So the job becomes visual. Someone watches. Someone captures. Someone types into a sheet if they have time. Google Lens on a screenshot is the "automated" version of the same ritual. You still chose which frame. You still missed the sticker that appeared at 00:07.
Food accounts, event promoters, and retailers all do this to each other. The intern's folder is the dataset.
I do not blame the folder. Until you have a row, a screenshot is the only artifact that survives a For You pass you will never see again. The problem is what happens next. Nobody searches a folder. Nobody diffs this week’s overlay against last week’s. The PNG is proof you watched, not a field you can filter for "20%" or a city name. Stickers make it worse: they are native objects on the video, not burned into a caption you already scrape, and they are the first thing a pause-and-capture workflow drops when the editor is late.
Do not collect frames. Collect strings.
I built the TikTok Video Text Extractor API so you paste public video URLs and get the on-screen strings back: overlays, captions, stickers, watermarks, hashtags, mentions, each with a timestamp when it first shows, a position if it holds still, and a confidence that is honest about blur and motion. No login. Speech is excluded. If they said the price and never showed it, you will not get the price from this API — and if they showed it and never said it, a transcript will not save you.
How it runs:
- Paste TikTok video links (standard or short links).
- Each public file is read for visible type.
- One row per video, segments you can filter, not a PNG named
final-final-2.
Who this is for
If you build competitor decks from ShopTok and you are tired of the screenshot ritual, this is for you. If you index promotional text for retail or hospitality research, this is for you. If you care about hook lines on screen in the first second, this is for you.
If you need comments, DMs, or private accounts, no. I only want the type a public viewer already saw.
Deleted videos skip. Very large files skip. Fast or tiny type is weaker. Volume can get throttled. Input, pricing, and limits are on the Apify Store README for TikTok Video Text Extractor.
A folder of screenshots is not monitoring
The cost of the old way is a price you saw on Tuesday that you cannot find on Friday because the PNG was named after the creator, not the offer. Or a report that undercounts codes because nobody wanted to pause fifty videos.
Stop screenshotting the overlay. Read it.
YouTube does the same trick with burned-in cards — extract on-screen text from YouTube Shorts when the channel posted there instead. The wider catalog is the all posts.