Working in your browser

Image to Text

Extract the text from a photo, screenshot or scan - right in your browser. Nothing is uploaded.

Local processing0 uploads~1.2s avgNo account required

Nothing is uploaded

The OCR engine runs inside this page. Your ID, invoice or contract never leaves the device.

Small images get enlarged

Undersized screenshots are scaled up to the letter height the engine needs, instead of being handed back to you.

Several languages at once

Read a mixed-script screenshot in one pass instead of picking a single language and losing half of it.

Up to 20 images

Process a whole batch and download every result together. No account, no daily cap.

In short: To extract text from image files, drop in a photo or screenshot, pick the language or languages it contains, and get the text back. Everything happens in your browser, so nothing is uploaded. If the picture is a scan or was taken at an angle, leave the cleanup options on - they matter more than which tool you use.

Why the same photo gives different results on different sites

Here is something the marketing pages do not mention: most free tools that extract text from image files are running the same open-source engine, Tesseract, compiled to WebAssembly. We checked ten of them and at least five say so outright. So when one site gives you clean text and another gives you nonsense from the same picture, the engine is rarely the variable.

What differs in any image to text conversion is what reaches the engine. And the single factor that matters most is one nobody advertises: how many pixels tall the letters are. Tesseract wants roughly 20 pixels of height per lower-case letter. A screenshot that has been forwarded through a couple of chat apps, or a photo scaled down for upload, often arrives at half that - and accuracy falls off a cliff.

What the other tools tell you to do

Retake it. Of the ten we tested, the usual advice for a poor result is to re-photograph or re-screenshot at a higher resolution. That is sound advice and completely useless when the image is the only one you have.

This page enlarges an undersized image before reading it instead, and says so when it has. In testing, a 400 by 184 pixel screenshot - small enough that most tools would tell you to start over - came back at 94% confidence with every line correct. Enlarging cannot invent detail that was never captured, but it gives the engine the letter height it was built for.

What OCR actually does, and what it does not

Optical character recognition looks at the pattern of dark and light pixels in an image and works out which characters those shapes most likely represent. It does not read or understand the page - it matches shapes to letters, one at a time.

That distinction explains almost every complaint people have about OCR. It is why a word in an unusual font comes out wrong, why a letter behind a coffee stain becomes a different letter, and why the engine will happily produce a confident-looking word that does not exist. It is pattern matching, not comprehension.

It also explains why every tool that can extract text from image data reports a confidence score for every single word it outputs. Those scores are real information, and this page shows them: when the average drops below 70%, the result is flagged rather than handed to you as if it were certain. Most tools hide this.

What people actually use it for

Screenshots and error messages

Copying text out of a screenshot someone sent you, or out of an error dialog that refuses to be selected.

Receipts and invoices

Pulling amounts and reference numbers off a photographed receipt instead of retyping them into a spreadsheet.

Photographed pages

Turning a picture of a printed page, a book, or a form into text you can search, edit and paste.

Documents you should not upload

IDs, medical letters, contracts. These are exactly the images most worth keeping off someone else's server.

How to extract text from an image

You can extract text from image files in three steps. No installation, no account, and no upload step. The first run downloads the OCR engine and your chosen language data into the browser cache; after that it works offline.

  1. 1

    Add your images

    Drag in a photo, screenshot or scan - JPG, PNG, WebP, BMP and GIF all work. You can add up to 20 at once and they will be processed one after another.

  2. 2

    Pick the languages and cleanup

    Select every language that appears in the image, not just the main one. Leave the cleanup option on for scans and photos; turn it off for screenshots, which are already clean. Turn on straightening only if the page was shot at an angle.

  3. 3

    Extract, then check and copy

    The text appears in an editable box per image, with a confidence score. Copy it, download it as a .txt file, or grab everything as a ZIP. Always skim the result against the original - OCR output looks confident even when it is wrong.

The engine and language data are cached after the first run, so the second image you process starts reading immediately.

The three things that decide whether it works

When you extract text from image captures, a poor result is almost always down to one of these - and the first one matters more than the other two combined.

Resolution, by a wide margin

Tesseract wants the lower-case letters to be roughly 20 pixels tall, which on a full page of text means the image should be around a thousand pixels on its short side. Below that, accuracy drops sharply. This page enlarges undersized images before reading them and tells you when it has done so - in testing, a 400 by 184 screenshot still came back with every line correct. That said, enlarging a blurry photo only gives you a bigger blurry photo. If you can retake the picture, fill the frame with the page: that beats every setting on this or any other site.

Contrast, and why cleanup helps less than you would think

The engine needs to separate ink from paper, so a grey scan or a shadow across half the page makes life harder. The cleanup option converts the image to black and white using a local adaptive threshold, computed per region rather than once for the whole picture, which is what a photo with uneven lighting needs. Being straight about the size of the effect: modern Tesseract already does its own binarisation internally, so on a reasonably lit photo the difference is small. In our tests it moved confidence by a few points and occasionally added a stray character. It earns its place on genuinely grey scans and low-contrast phone photos, not on screenshots.

Angle, for photographed pages only

OCR reads in horizontal lines, so a page tilted by several degrees means words start drifting across the rows the engine is scanning. Straightening detects the dominant text angle and rotates the image back first. Turn it on for anything photographed by hand at an angle. Leave it off for screenshots and flatbed scans - there is no tilt to correct, and the detection can only introduce error.

The honest summary: start with the largest, sharpest version of the image you have. Then use cleanup if it is a scan or a dim photo, and straightening if it was shot at an angle. Neither will rescue an image that was too small to begin with.

Screenshots with more than one language - the one thing no other tool does

This one catches people out. Every tool we tested asks you to choose a single language before extracting. That is fine for a page of English, but a great many real images are not that: a Chinese invoice with English product names, a Japanese interface with Latin brand names, a Russian document quoting an English URL.

Pick one language and any attempt to extract text from image regions in the other script does poorly - not because it cannot see those characters, but because it is matching them against the wrong model. You get half your text and a lot of nonsense where the other half should be.

Tesseract can load several language models at once, and this page lets you select as many as you need. Of the ten tools we tested, not one offers this: every single one makes you commit to a single language before extracting. The trade-off is honest and worth knowing - each language adds to the data downloaded on first use, and loading several makes recognition somewhat slower. Select the ones actually present in your image rather than everything on the list.

What this will not do

OCR has real limits, and knowing them before you extract text from image files saves more time than any setting.

Handwriting, especially cursive

Tesseract is trained on printed type. Neat block capitals sometimes come through; ordinary joined-up handwriting generally does not. This is a limitation of the engine, not of this page, and it applies to every free tool using it - though most do not say so.

Formatting and layout

You get plain text. Bold, italics, headings, colour and font are all discarded. Tables lose their structure and become rows of words, and a two-column page may be read straight across instead of column by column.

Rescuing a genuinely bad photo

Cleanup helps a mediocre image. It cannot invent detail that was never captured - a blurry, dim or very small photo of small print will not become readable. Retaking it at a higher resolution beats any amount of processing.

Understanding the document

It returns the characters it sees, in reading order, and nothing more. It does not know which number is the total or which line is the date. Check anything you are going to act on.

How the free tools compare

Ten tools that extract text from image uploads, checked in September 2026 by using each one. Free-tier behaviour only, and only the things that vary - all of them are free and most run in the browser.

ToolWhere it runsBatchSeveral languages at once
This pageYour browserUp to 20Yes - pick any combination
PixConvertYour browserOne at a timeNo - one at a time
UtiloKitYour browserOne at a timeNo - one at a time
OpenConvertYour browserYesNo - one at a time
PixellizeYour browserOne at a timeNo - one at a time
imagetotext.infoNot statedPaid tierNo - one at a time
OnlineOCR.netTheir serverSigned-in onlyNo - one at a time

The honest summary is that several of these image to text tools are perfectly good, and the ones running in your browser deserve credit for it - that used to be the differentiator and no longer is. Two things do still vary: whether you can hand it a whole batch, and whether you can read a mixed-script image in one pass. On the second one, we did not find another tool that lets you.

All tools are 100% free and secure

Complete PDF Tool Suite

Discover our comprehensive collection of PDF tools designed to handle all your document needs

Index of tools · 47
01

PNG to PDF

Bind PNG images into a single, print-ready PDF

Avg 1.2sLocal
02

JPG to PDF

Convert JPG images to PDF format

Avg 1.4sLocal
03

Merge PDF

Combine multiple PDF files into one

Avg 0.8sLocal
04

Compress PDF

Reduce PDF file size efficiently

Avg 2.1sLocal
05

PDF to PNG

Convert PDF pages to PNG images

Avg 1.6sLocal
06

PDF to JPG

Convert PDF pages to JPG images

Avg 1.5sLocal
07

PDF to Text

Extract text content from PDF files

Avg 0.9sLocal
08

Split PDF

Split PDF into separate pages

Avg 0.6sLocal
09

Edit PDF

Edit and annotate PDF documents

BetaLocal
10

Organize PDF

Organize and rearrange PDF pages

Avg 0.7sLocal
11

Rotate PDF

Rotate PDF pages and save permanently

Avg 0.5sLocal
12

Page Numbers

Add page numbers to PDF with live preview

Avg 0.6sLocal
13

Watermark PDF

Add a text watermark to PDF with live preview

Avg 0.6sLocal
14

HEIC to JPG

Convert iPhone HEIC photos to JPG

Avg 1.0sLocal
15

Delete PDF Pages

Remove unwanted pages and download a clean PDF

Avg 0.5sLocal
16

Extract PDF Pages

Save selected pages as a new PDF or separate files

Avg 0.5sLocal
17

Sign PDF

Draw, type, or upload a signature and place it on any page

Avg 1.0sLocal
18

Resize PDF

Change page size to A4, Letter, or custom dimensions

Avg 0.6sLocal
19

Crop PDF

Trim margins, drag-select, or auto-crop white space

Avg 0.6sLocal
20

Flatten PDF

Make forms read-only — keep text searchable or lock to image

Avg 0.7sLocal
21

PDF Metadata

View, edit, or strip author, title, dates and other metadata

Avg 0.4sLocal
22

Grayscale PDF

Convert to grayscale or black & white to save colour ink

Avg 1.8sLocal
23

Extract Images from PDF

Pull embedded photos out of a PDF and save as PNG or JPG

Avg 0.9sLocal
24

WebP to PDF

Convert WebP images to a PDF, merge many into one file

Avg 0.7sLocal
25

Protect PDF

Add an open password to your PDF, entirely in your browser

Avg 0.3sLocal
26

Unlock PDF

Remove a known password or restrictions from a PDF, in your browser

Avg 1.5sLocal
27

HEIC to PDF

Convert iPhone HEIC photos to PDF

Avg 0.8sLocal
28

AVIF to PDF

Convert AVIF images to PDF

Avg 0.7sLocal
29

TIFF to PDF

Convert TIFF images and scans to PDF

Avg 1.0sLocal
30

BMP to PDF

Convert BMP bitmap images to PDF

Avg 0.7sLocal
31

OCR PDF

Make a scanned PDF searchable, entirely in your browser

Avg 2s/pageLocal
32

Redact PDF

Permanently remove sensitive text from a PDF, in your browser

Avg 0.3s/pageLocal
33

Compress Image

Shrink JPEG, PNG and WebP in batch, entirely in your browser

Avg 0.2sLocal
34

Resize Image

Change image dimensions in batch, with presets, in your browser

Avg 0.2sLocal
35

PNG to JPG

Turn PNG images into JPG, batch, with the size shown before you download

Avg 0.2sLocal
36

JPG to PNG

Turn JPG images into PNG, batch, with optional transparent background

Avg 0.3sLocal
37

WebP to JPG

Turn WebP files into JPG, batch, with every frame of an animated WebP

Avg 0.3sLocal
38

WebP to PNG

Turn WebP into PNG with transparency intact, or knock a solid background out

Avg 0.3sLocal
39

Image to WebP

Turn JPG, PNG and more into WebP, batch, or hit an exact size in KB

Avg 0.2sLocal
40

SVG to PNG

Rasterise SVG at any size, or make a whole favicon set from one logo

Avg 0.2sLocal
41

AVIF to JPG

Turn AVIF images into JPG so they open anywhere, or hit an exact size in KB

Avg 0.3sLocal
42

Compare PDF

See exactly what changed between two PDF versions, without uploading either

Two files, no uploadLocal
43

SVG to JPG

Rasterise SVG to JPG at any size, and choose what colour replaces the transparency

Pick size and backgroundLocal
44

GIF to PNG

Take one still frame, or split an animated GIF into a numbered PNG per frame

Every frame at onceLocal
45

GIF to JPG

Choose what replaces transparency, take one frame or all - and we say when PNG is the better target

You pick the backgroundLocal
46

PDF to TIFF

Turn a PDF into one multi-page TIFF

One multi-page fileLocal
47

Image to Text

Extract text from photos and screenshots

11 languages, offlineLocal

Questions about extracting text from images

What people ask before they extract text from image files and rely on the result.

Is it free, and are there limits?

Free, with no account, no daily cap and no watermark. You can process 20 images at a time, each up to 25 MB. That limit exists because OCR is CPU-heavy and runs on your own machine - not because of a paid tier. Run another batch straight after.

Are my images uploaded anywhere?

No. The OCR engine runs inside this page. Your images are read by your own browser and never transmitted. This matters more here than for most conversions, because the pictures people run through OCR are IDs, invoices, medical letters and contracts.

Why did it download several megabytes the first time?

That is the OCR engine and the language data, and it only happens once. Because the recognition runs locally rather than on a server, the engine has to reach your device. It is cached afterwards, so later runs start immediately - and work offline.

Can it read handwriting?

Not reliably, and cursive usually not at all. The engine is trained on printed type. Careful block capitals sometimes come through readably. This is true of every free tool built on Tesseract, which is most of them.

The text came out wrong. What should I change?

Resolution first, then cleanup, then the angle. When you extract text from image files, a larger or sharper source beats every setting - it is not close. After that, turn cleanup on if it is a grey scan or a dim photo, and straightening on if the page was shot at a tilt. If your image has two scripts in it, select both languages rather than guessing which one dominates.

Can it handle an image with two languages in it?

Yes - select every language present, not just the main one. The engine loads several models at once and reads both scripts in a single pass. Most other tools make you choose exactly one, which is why mixed-script screenshots come back half-empty from them.

Does it keep the layout, tables or bold text?

No. The output is plain text. Formatting, colour and font are discarded, tables lose their structure, and multi-column pages may be read across rather than down. If you need the layout preserved, you want a PDF with a text layer rather than extracted text.

What if I have a PDF rather than an image?

Use the PDF OCR tool instead - it keeps the pages intact. It renders each page, reads it, and writes an invisible text layer back over the original so the document looks unchanged but becomes searchable. Converting a PDF to images first and running them through here would lose the document structure.

Related tools

Other tools that pair well when you extract text from image files, all free and all running on your own device.

  • OCR a PDF for documents rather than pictures - adds a searchable text layer and keeps the pages intact.
  • PDF to text when the PDF already has real text in it and just needs extracting, no OCR required.
  • Resize an image enlarge an undersized screenshot before reading it, or shrink an oversized photo.
  • Compress an image for when the picture is fine but the file is too big to send on.
  • PDF to PNG pull pages out of a PDF as images, if you need the picture rather than the words.