Extract Text from Image Privacy Policy

Extract Text from Image pulls the text out of pictures, screenshots, and scanned PDF pages. Almost everything happens inside your own browser. This page explains, in plain words, the few moments when anything is sent to us.

On-device recognition

Pictures, screenshots, and PDF files are read on your device for OCR. They are not uploaded for recognition. PDF passwords stay on your device. Clean also runs on your device, and the address of the page a picture came from is not saved. If you choose Share source, the selected image is encrypted on your device before upload; the Share section explains what is sent.

The features that go online

Summary and Translate send only the text already recognized on your screen. Opening them sends nothing. A request goes out only after you press Generate summary or pick a language. It carries only that text over an encrypted connection. Your picture, your PDF file, and your PDF password are never part of it.

Our service reads that text, writes the summary or the translation, sends it back, and keeps no copy. Our records of these requests hold only ordinary service details, never your text.

Some networks cannot reach our service at all. Your request may then take a second route through another server we run, which passes it along unchanged and hands the answer back. That server stores nothing at all.

Share is optional. You can share a .txt, .md, or .docx export, or a PNG source image. The extension encrypts the selected content on your device with AES-GCM and uploads only ciphertext. The decryption key stays after the # in the link and is not sent to our service. The link stops working after 15 minutes. The KV access token also expires after 15 minutes. In the current Worker configuration, R2 cleanup runs daily and removes objects older than 30 minutes, so encrypted bytes may remain for up to about 24 hours 30 minutes after upload. Anyone with the complete link can download the file only while the link works, so share it only with the person you intend.

If you switch on Chinese, Japanese, or Korean, the extension downloads that language's recognition files once from our service so it can read that script on your device. The request asks for a file and sends nothing about you or your images; afterwards that language works offline too.

There is a daily cap on how many summaries and translations we handle in total, to keep costs predictable. If it is reached, both features pause until the next day. Recognition and Clean keep working, because they never needed the internet.

The number sent with those requests

Your copy of the extension invents a random number for itself and keeps it on your device. It travels with service and product-analytics requests so fair-use limits count per copy of the extension instead of per network, and so we can count active installations and understand which features work or fail.

It is not an account or an email address, and it is never tied to you or to what you sent. Remove the extension or clear its data and it is gone; a fresh install makes a new one.

Anonymous product analytics

The extension sends small first-party events when it is installed or updated, opened, completes or fails a feature, creates an export, or when you use Share file, Copy link, or QR. These events can include the extension version, broad feature name, success or error category, export format, interface entry point, language code, operation time and approximate byte or item count. They never include your image, PDF, recognized text, summary, translation, filename, page address, Share link, Share token, or anything from your clipboard.

Each event carries the random installation number described above, a short-lived random session number, a random event number that prevents retry duplicates, and, when present, a random acquisition number used to connect the install redirect with the Welcome page. Our Worker checks these values and stores only one-way hashes of the installation, session, acquisition and temporary Share identifiers. The random event number is stored only for deduplication. Cloudflare derives the request country at the edge; we do not store the raw IP address or full browser User-Agent in product analytics.

We use these events to measure active installations, feature adoption, reliability, retention, versions, broad geography, and the Share/Copy/QR funnel. They are not used for advertising, cross-site tracking, credit decisions, or personal profiling. Cloudflare Analytics Engine keeps event data for up to three months; aggregate reports without identifiers may be kept longer.

Install and removal analytics

The welcome page and the hosted goodbye and rating pages carry analytics used to count installs and removals by country.

Optional feedback

After you remove the extension, you can share optional feedback on the goodbye page. If you rate the extension one to three stars, you can add a comment on the rating page.

What stays on your device

If you leave History on, the extension keeps recent results, the text, a small picture, and the file name in your browser only. It holds the 50 newest unpinned results and uses a 25 MB logical budget based on UTF-8 text, thumbnails, file names, and a per-record metadata reserve. Older unpinned results drop off on their own, and anything you pin stays. You can delete one result, clear everything, or switch History off in Settings.

Your settings stay on your device too. There are no accounts, advertising identifiers, third-party advertising trackers, or session-replay tools inside the extension.

What we never do

We do not sell your data, share it for advertising, or show you ads. The hosting company whose computers run our service sees Summary and Translate text only while a request is being answered and does not train on it. For Share, it receives encrypted bytes without the decryption key. The link and access token expire after 15 minutes; scheduled cleanup of the encrypted bytes may take up to about 24 hours 30 minutes in the current Worker configuration. It also processes the anonymous product events described above. Those computers sit in several countries, so this processing may happen outside your own country.

Your rights

We use your text only to provide the feature you requested. You can stop using Summary, Translate, or Share at any time. Wherever European, UK, or California privacy law applies, you may ask what we hold about you, ask for a copy, ask us to correct or delete it, ask us to limit a use, and complain to your privacy authority. Email support to make a request. Shared files are encrypted. Their links stop working after 15 minutes, while encrypted bytes may remain until scheduled cleanup.

Limited Use

The use of information received from Google APIs will adhere to the Chrome Web Store User Data Policy, including the Limited Use requirements.

Children

This is a general tool for work and study. It is not made for children under 13, and we do not knowingly collect anything from them.

Changes

If this policy changes, the new version is posted on this page and the date below changes with it.

Contact

Questions, or want to use one of the rights above? Email support.

Last updated: September 28, 2026