Live browser demo
Find visible text in a screenshot
Upload an image and get a point back, computed locally.
This live OCR demo is not the GUI-Actor model in the recordings. No accuracy benchmark has been published for this OCR example.
A small OCR model reads the image. We match the visible text you specify and return its center as numerical coordinates. This is a practical text-based alternative; it does not run GUI-Actor or understand arbitrary instructions.
Limits and privacy
Tesseract.js 5.1.1 runs an English text-recognition model on this device after the first download. Your screenshot is not uploaded. Images are reduced to at most 1,600 pixels on their longest side for inference; coordinates are mapped back to the original dimensions.
Exactly one text match returns a point. Multiple matches return AMBIGUOUS; no match returns NO TEXT MATCH. Neither proves a button is absent or that clicking the text is safe. Icons, tiny text and complex layouts may fail. The demo never sends a click.