BlurMate

Home / Image to text

Image to text

Reads the text in a photo or a screenshot so you do not have to type it again. Fix whatever came out wrong, then copy it or save it as a TXT file. The image is never uploaded — the reading happens in this browser tab.

Choose an image and its size and format are checked first.

English, Korean, JapaneseRotate before reading

For the things you would otherwise retype

Reading text out of an image turns a picture of words into words you can edit. The receipts that have to go into an expense claim. The whiteboard somebody photographed before the room was needed. An error dialog that will not come back so you can select the message. A page of a book, a business card, a slide from a talk that was never handed out. All of it arrives at the same place: you have the words, but only as pixels, and typing them out again is twenty minutes you did not plan on.

The reading program is downloaded from this site and runs inside this browser tab, so the image itself stays on your device. That matters when the paper in the photo is a contract, a payslip or an ID — the sort of thing you would not paste into a website that offers to read it for you. If you want to confirm it rather than take our word for it, open your browser's developer tools, watch the Network panel and press Read text: you will see the program and the language data come down, and nothing go up.

Which languages it can read

There are three choices, and they are the whole list: English, Korean + English, and Japanese + English. Those are the only three sets of language data kept on this site. Pick the one that matches the writing in your image. The two combined options read English in the same pass, so a Korean or Japanese document with English words and numbers mixed into it does not need a second run.

There is no data here for anything else. Chinese, Arabic, Hebrew, Thai and the rest are not available, and there is no right-to-left reading. A page in French, Spanish or German will be read as though it were English: the letter shapes it knows come through, and accented characters and unfamiliar words become guesses. That is worth knowing before you judge the result — a bad read can mean the wrong language was picked rather than a bad photo.

The download happens on the first read, not when the page loads. English on its own is about 6MB, Korean + English about 7MB, and Japanese + English about 8MB; the note under the language picker always shows the size for the option you have selected. It is the program and the language data coming to you, and it goes one way only.

What reads cleanly, and what does not

The reader works line by line on shapes it has been trained on. It does best with printed type, sitting level, dark on light, photographed straight on under even light — a scan, a screenshot, a flat page under a window. It does worst with handwriting, with a page shot at an angle from across a desk, with a screenshot that was already shrunk before you got it, and with words printed over a photo or a pattern.

Angle is the one worth checking before you press the button. In our own test images — Korean lines plus a line of English and numbers, drawn into a picture rather than photographed, read with the Korean + English option on 2026-08-29 — a 5° tilt still came out right, while the same lines rotated 90° produced thirty lines of nonsense. A phone photo saved sideways falls into that second case, so turn it upright with the Left and Right buttons first. Grey text on a grey background was read correctly in that test but took longer than anything else we tried, which is the practical argument for shooting the page in better light instead of in shade.

One more result from the same test: a photo of a car with no writing in it returned ten lines of nonsense. The reader does not tell you it found nothing; it offers its best guess at what the shapes might be. So if the output looks like scrambled letters, the question is not how to improve the settings — it is whether there is legible text in the image at all, and whether the right language is selected. Zoom the preview between 50% and 300% and read the words yourself before you decide.

Where people get caught out

If I can read it, the tool can

Your eye fills in what the reader cannot. You know what the word on the crumpled receipt says because you know what receipts say. The reader has the pixels and nothing else, so contrast, letter size and a level line matter far more than they do to you. When a read comes back poor, the fix is usually a better picture — closer, straighter, brighter — rather than another attempt at the same one.

The layout comes across as well

Tables and two-column pages come out in the wrong order. Text at the same height across the page can be joined into a single line instead of being kept in its column, so a table turns into rows that mix cells and a two-column article interleaves its columns. What you get back is plain text — no columns, no bold, no cell boundaries. For a table, crop and read it a column or a cell at a time, or put the lines back in order in the result box.

Numbers are the safe part

Digits are where the quiet mistakes live. Zero and capital O, one and lowercase l, five and S all trade places, and unlike a misread word a wrong digit reads perfectly plausibly. Account numbers, reference codes, dates and totals are worth checking against the original one character at a time. The character count above the result box gives you a quick way to tell whether whole chunks went missing.

The result only exists in this tab. Close the page and it is gone, so fix it and then copy it or download it — the TXT file is named after your image, with -text.txt on the end. And the reader covers what it can recognise, not everything a person could: check the text against the picture before you use it somewhere it matters.