Read text from an image
The AI reads the text in a photo, a screenshot or a scan and turns it into text you can edit. Correct it, then copy it or save it as a text file. The image is read only on this device.
- Free
- No upload
- No sign-up
- Works offline
How to use
Press "Choose a file" and pick one image with text. The AI reads its text.
Compare the text with the image. You can correct any misread text right there.
Press "Copy" to paste the text into another app, or press "Save" to save it as a text file (text.txt).
Questions
Is my image sent anywhere?
No. The AI that reads the text runs on your device too. The first time, the parts of the AI (about 43 MB) are loaded from the TEMOTO site, but nothing is sent out after you choose an image.
Which languages can it read?
Japanese (hiragana, katakana and kanji, vertical text too) and English. It also reads Chinese characters and letters with accents (French, German and others), but only short texts of them were tested. It cannot read Korean, Russian, Thai, Arabic or other scripts. When such characters are inside a line of Japanese or English, they may be left out or read as other characters, without a warning.
How accurate is it?
In the test images, about 4 in 1,000 characters of printed horizontal text were misread, and about 16 in 1,000 of vertical Japanese. Very small or blurry text is misread more often (about 17 in 100 characters for letters 6 pixels high). Compare the text with the image before you use it.
Can it read handwriting?
It was not tested with real handwriting. Handwritten text may be misread more often, or not read at all.
What about tables and text in columns?
The text is read, but lines can come out in the wrong order. Text side by side, such as the items and prices of a receipt, is read as one line. Check the order after reading.
Can it read large images?
An image with a longer side over 3,200 pixels is made smaller to 3,200 pixels before it is read. If small text was missed, try an image cropped to the part you want to read.
Can I use it on a phone?
Yes. It needs WebAssembly SIMD, so use a recent version of Chrome, Edge or Firefox, or Safari 16.4 or later. The first time, the parts of the AI are loaded, so Wi-Fi is a good idea. Inside apps such as LINE, saving may not work. If so, open the page again in Safari or Chrome.
Limits
- The AI can misread or miss characters. Compare the text with the image before you use it.
- Handwriting, very small text and blurry text are misread more often. In tables and columns, lines can come out in the wrong order.
- Sideways or upside-down images cannot be read as they are. When the tool finds that an image is turned, it says which way to turn it with "Rotate and flip images".
- Each time, one image can be read. An image with a longer side over 3,200 pixels is made smaller first. For moving pictures (GIF, moving WebP or PNG), only the first frame is read.
- One file can be up to 50 MB, and a picture up to 50 megapixels. HEIC (the photo format of iPhones) works only in browsers that can open it (mainly Safari).
- The first time, the parts of the AI (about 43 MB) are loaded. After that, the parts kept on this device are usually used.
- We cannot protect against things in your own environment, such as browser extensions.