Text Recognition: the guide

Pull copyable text out of images with a local OCR engine - the image never leaves your device. Entry point: https://lab.stsqz.com/ocr.

How to use it

1. Upload an image containing text (screenshots work best)
2. The engine finds the text regions automatically and recognizes them block by block
3. Results appear in region order, with one-click copy of the whole text

About the engine

It runs PP-OCRv4 in two stages: a detection model boxes every text region in the image, then a recognition model turns each box into text. First use downloads about 16MB in total; after that it works offline. Very large images are scaled down proportionally to the engine's working size before recognition.

Good use cases

• Code, error messages, and chat logs inside screenshots
• Whole passages from book pages and exam papers
• Mixed Chinese-and-English text is recognized too

Improving accuracy

• Use sharp images; heavy compression artifacts get in the way
• Keep text horizontal - rotate flipped images first
• For very small type, upscale the image before uploading
• On complex layouts (tables, multi-column), blocks may be ordered slightly differently from how you'd read them; a quick manual re-order after pasting is fastest

Playing well with other tools

Paste the recognized text straight into the Text Tools for counting or format conversion, or hand it to piko's brain for a summary - a tidy all-local pipeline.