OCR here is Tesseract.js. The engine, WASM and traineddata load from public CDNs (cdnjs, then jsDelivr, then unpkg). We do not run recognition on a GPU farm of our own — your pixels stay in the tab.
When to skip OCR
If you only need the scan inside a document, use Place image. It is instant, works offline, and cannot garble a name. OCR is for text you plan to copy and edit.
Limits
Eight images in the list. JPEG, PNG and WebP. Handwriting and low-contrast photos will miss letters; that is the engine, not a server timeout.
Need the reverse direction? wordtojpg.net snapshots a .docx to JPEG pages.