Adds an invisible, searchable text layer to a PDF and re-exports
it — the scanned pages themselves are untouched. A plain-text copy of
what was read comes with every job.
Choose a file, or drop it hereOne PDF — or several page images, combined into one document
PDF, PNG, JPEG, TIFF, BMP or WebP.
Rasterising discards the original page image. Use
“replace” to keep the scan intact. Surya ignores this and
the language list — it detects script itself, and always writes a
fresh layer.
Ctrl-click (⌘ on Mac) to pick several. Put the
dominant language first — more languages is slower and slightly
noisier.
Changes only what the engine sees, never the saved page.
Used only to decide between the two
engines when they disagree. Pick the document’s language, or
None — a wrong dictionary rejects every word and the arbiter
stops deciding rather than deciding badly.
A spread OCRed whole gives each page half the
resolution. Surya only.
Glyphless matches pre-v17 output, for the OCR-layer
editor and the corrections applier.
Runs a 4B model on CPU — budget around four
minutes per page, or six if both engines run. Jobs can be
cancelled.