What this guide covers
A useful region contains the signal you want and very little unrelated content. Large captures add competing text, animation and processing work. For a price alert, capture the price rather than the entire product page; for a status word, include enough margin that the word is not cut off. The region follows image coordinates, not a webpage element or its underlying document.
Set it up
Connect the intended source and open the region picker. Drag around the text, number or visual target, leaving a small clear border. Save the task and compare its next result with the preview. If the application moves a panel, changes resolution or changes its layout, reopen the picker instead of assuming the original coordinates still describe the same content.
Check the result
For text, keep whole character shapes and avoid clipping ascenders, descenders or punctuation. For a number, exclude adjacent dates, stock counts and labels containing other digits: numeric rules use the first parsed number. For an image match, use an area whose dimensions and aspect ratio resemble the reference pattern rather than stretching a large unrelated scene.
Before you rely on it
Browser zoom and responsive layouts can relocate content even when the window title remains unchanged. Scrolling a shared webpage also changes what is inside a coordinate-based region. Test the expected source arrangement, then leave it stable. Watch does not pin a DOM element, fetch a hidden webpage or automatically scroll the source back to its former position.
Troubleshooting
If results alternate between plausible and unrelated text, inspect whether a tooltip, advertisement or popup enters the region. If the crop is clear but empty, confirm that the task belongs to the connected source and has not been paused. Reduce background motion and increase the source font size before widening the crop to include an entire screen.
Product reference
A tight crop also limits which recognized words can enter records or selected external actions, but the authorized capture stream still comes from the source selected in the browser. Treat region selection as recognition setup rather than an independent operating-system privacy barrier. Review both choices before using personal or work information.
Add a source in OKYA Watch and authorize the window or tab you need. Create a text recognition task, select the text region and OCR language, then save and start recognition. Copy the result or export recognition records as CSV. It reads visible text in the shared image, not webpage source code or document files.
Further reading
- Local OCR models and recognition languages
- Number changes and threshold alerts
- Set up system, Webhook, Telegram and email notifications