Pick an image
Drop a screenshot, a scan or a photo onto the page, or choose a file from your device.
Extract Text from Image Online Free
Turn a picture into editable text with PP-OCRv6, a compact OCR model that runs entirely in your browser. The model is fetched once from Hugging Face and cached, so your photo never leaves the device.
Pick an image, then load the model. Extraction reuses the loaded model.
The medium model pair is about 140 MB (detection + recognition) plus a dictionary. Files come from a public model hub, not from 73hi, and your browser caches them for the next visit. No image is uploaded at any point.
Your image appears here
0 lines
Three simple steps
How to Extract Text from an Image Online
Read the words inside a picture without an account, a queue or a server round trip.
Load the model
Select Load model and wait until PP-OCRv6 is ready, then select Extract text. Extraction uses the loaded files and never downloads the model again.
Copy or download
Detected lines show as boxes on the picture and as editable text you can copy to the clipboard or save as a .txt file.
How it works
A Compact OCR Model, Entirely on Your Device
The whole pipeline runs in the page: the picture is scanned for text regions, each region is cropped and fed to a recognition network, and the characters are decoded.
| Stage | What it does | Where it runs |
|---|---|---|
| Text detection | Finds every text region in the image as a polygon | Your browser |
| Crop and normalise | Slices each region and prepares it for the recogniser | Your browser |
| Text recognition | Reads the characters with a CTC decoder and a dictionary | Your browser |
One time cost: the model files total about 140 MB for the medium pair. They come from a public model hub, not from 73hi, and your browser caches them for the next visit.
Questions and answers
Image Text Extractor FAQ
Where does the model come from?
On 73hi.com the page downloads PP-OCRv6 from Hugging Face, and on 73hi.net from ModelScope. Both serve the files with permissive cross origin headers, and the 73hi server never stores or forwards the model.
Is my image uploaded?
No. The picture is decoded and recognised entirely on your device. Only the model files are downloaded, and they are cached by the browser after the first run.
Which languages does it read?
PP-OCRv6 is trained for mixed Chinese and English plus digits and common symbols, so it handles English, Chinese and mixed-script text on the same page.
Do I need a GPU?
No. The model runs on the CPU through WebAssembly. This page uses WebGPU automatically when your browser supports it (which makes recognition faster) and falls back to the CPU when it does not. A GPU is optional.
Why is the first load slow?
The medium model pair is about 140 MB (detection plus recognition) plus a dictionary, so the first download can take a while on a slow connection. Smaller presets download and run faster with somewhat lower accuracy.
Can I use files I already downloaded?
The model is pulled straight from the public hub and cached by the browser, so there is no separate local-file option on this page. Clearing the browser cache forces a fresh download on the next visit.