Shot2Text

About Shot2Text ? Who Runs It and Why It Stays Local

Shot2Text is a small independent project, not a company. Here is why the recognition never leaves your browser, what we refuse to build, and where this tool is genuinely weak.

Who runs this site

Shot2Text is a small independent project, not a company with a support department. One person builds and maintains it. There is no funding round, no growth team, and nobody whose job is to write blog posts for search engines.

That matters for how you read everything else here. There is no commercial pressure to invent features we do not have, no reason to hide the limitations, and no sales funnel behind the tool. If something does not work, the honest answer is usually more useful to both of us than a marketing answer.

Why the recognition runs inside your browser

Every mainstream online OCR service works the same way: you upload an image, a server reads it, and the text comes back. That design is easier to build and easier to scale, and for a company it also means your document passes through their infrastructure.

This site does the opposite. The text reader is downloaded to your browser once — about 3 MB of English language data — and from then on the recognition happens on your own machine. Your image is never sent anywhere, because there is no server to send it to.

The reason is not that we are more virtuous than the other tools. It is that the trade-off looked clearly worth taking for this particular job:

  • Screenshots are ideal input. They are drawn by software, so the edges are exact and the contrast is flat. The accuracy gap between a browser reader and a paid server-side engine, which is real on difficult scans, mostly disappears on screenshots.
  • A screenshot is often more sensitive than it looks. An error message carries a project name. A receipt carries a card's last four digits. A chat screenshot carries someone else's words. Not uploading any of that is worth a little waiting.
  • No server means no per-image cost, which is what makes a genuinely free tool possible without ads.

What we will not build

Written down so that the answer to "will they ever add this?" does not depend on our mood.

We will not add accounts. No sign-up, no login, no profile, no "sync your history across devices". An account would mean storing who you are and what you converted, and the entire point here is that nothing is stored.

We will not upload your images, not for better accuracy, not for a premium tier, not "just to improve the service". If a future feature cannot work locally, it does not get built.

We will not add advertising. Ad scripts are third-party code running on the same page as your image, and they routinely load tracking pixels. That combination is incompatible with a promise about privacy. We would rather stay small.

We will not add watermarks or paid tiers. There is one version of each tool and it is the full version.

We will not add a public API. There is nothing server-side to expose. If you need OCR in an automated pipeline, the engine used here is Tesseract and it is open source — running it yourself is the honest recommendation, and it will be faster than calling anyone's endpoint.

Honest limits

These are the things this tool does not do well, stated plainly because you will find out anyway.

The language has to be chosen, and 18 are available. Each pack is 0.7 to 2.9 MB, so nothing is downloaded until you pick one — that is the whole reason the selector exists instead of loading everything. The catch is that it will not detect the language for you: read Chinese with English selected and you get nonsense rather than an error, so the dropdown is a step you have to remember.

Handwriting is unreliable. Neat, spaced-out printing converts reasonably well. Flowing cursive usually produces nonsense, because there are no clear letter boundaries to detect. No amount of retrying fixes it.

Tables and columns are not preserved. The reader works line by line across the full width of the image, so a two-column page comes out interleaved. Crop one column at a time if that matters.

There is no batch mode. One image at a time. Holding several in memory at once is what makes a phone browser tab slow down and close itself.

The first conversion takes a few seconds. That is the 3 MB language download. Later images take one or two seconds because the data stays cached.

There are no formatting controls. No output options, no export to Word from every tool, no choice of engine. Plain text and, on one page, a .docx.

How it is paid for

Right now it is not. There is no advertising, no subscription, no paid tier and no affiliate links, so the site earns nothing.

What makes that possible is the architecture rather than generosity. The whole site is static pages plus files that visitors download; there is no server doing per-image work, so there is no per-image bill. The ongoing cost is a domain name and a small hosting plan — the kind of number one person can absorb while a project is finding out whether anyone wants it.

If that ever stops being true, the fix will not be to start uploading your images or to put ads on the page. It will be to say so here.

How to get better results

Almost every disappointing conversion is an input problem. These are the changes that make the biggest difference, in order:

  1. Crop to the text. A full screenshot of a browser window is mostly interface. Cropping reduces errors and reduces cleanup.
  2. Use the sharpest file you still have. A PNG export beats a JPG screenshot, which beats the same image sent through a chat app.
  3. Shoot paper straight down, in even light. Tilt and your own shadow do more damage than any setting can repair.
  4. Zoom before capturing small text. Below roughly 12 pixels per character there is nothing left to recognise.
  5. Prefer light mode. Pale text on near-black is workable but noticeably weaker than dark on light.

The guide to extracting text from an image goes through each of these with the failure it prevents.

A note on the name

This site is shot2text.com. If the address in your browser is different, you are on a different tool — several services have similar names, and one in particular spells out "screenshot" in full.

Worth checking if you were recommended something by name: the address bar is the definitive answer, not the logo. We cannot speak for any other site's privacy policy, and it would be wrong of us to imply otherwise.

Contact

Bugs, a language you need, or a correction to something written here: contact@shot2text.com. Details are on the contact page.