Image to Text

Yes, but you have to flip one switch — light text on a dark background is the one case the engine cannot handle as-is, because it expects dark characters on a light page. Turn on Invert in the image clean-up panel and the screenshot is flipped to dark-on-light before recognition, which usually takes the result from unusable to clean. There is no auto-detection for this, deliberately: guessing wrong would silently ruin a normal screenshot, so it is left as a control you can see.

Resolution, nine times out of ten. OCR needs enough pixels per character to tell letterforms apart — roughly 25 pixels of character height, which is about what 300 DPI gives for normal body text. A photo of a document taken from too far away, or a screenshot scaled down before saving, simply does not contain the shapes any longer. Re-capture closer or at a higher resolution rather than trying different settings; sharpening a source that lacks the detail cannot recover it.

Not reliably, and it is fairer to say no. The engine is trained on printed and typeset characters, where letterforms are consistent, and handwriting breaks that assumption — joined letters, varying slant and personal letterforms produce output that takes longer to correct than retyping. Neat block capitals sometimes come through. Cursive and normal note-taking handwriting will not.

Because recognition is not purely shape matching — the engine uses the selected language's character set and word patterns to resolve ambiguous glyphs. With the wrong language selected, characters that exist in one alphabet and not another get forced into the nearest available shape, and the correction step then pulls words toward a vocabulary that does not apply. Selecting the actual language of the document is usually the single largest accuracy gain available, ahead of any image preprocessing.