FileFlow Guides

PRACTICAL FILE GUIDE

Scan to Word: correct OCR beside the original before exporting

A scan records how a page looks. OCR guesses which characters those pixels represent. Turning that guess into a useful document requires a review step, especially for names, dates and numbers. FileFlow keeps the original page beside the editable text so you can check each correction before downloading.

Updated:

Open the tool

Which document should you download?

FormatKeepsDoes not keep
DOCXEditable paragraphs, heading sizes and emphasisOriginal scanned layout, images and tables
TXTUnicode plain text and source-page separatorsBold, headings and list formatting
New PDFSelectable edited text, heading sizes and emphasisOriginal scan image or its exact geometry
PDF with scan and searchPage image and invisible OCR textManual word changes, forms and digital signatures

Choose the pages and recognition language first

Open PDFs or JPG, PNG and WebP images. The page list represents the source order; mark the pages to include and move them with the arrows. Up to 20 pages can be included in one export, so a long PDF should be handled in separate selections.

Choose the language of the printed document. The interface selects a matching recognition model when one is installed, otherwise English. Changing the interface does not translate the contents of your scan. A language-model download is needed for recognition, but your documents are not uploaded.

Correct the words while the original is visible

Recognize unfinished pages, then select a page to review it. Compare the original and editor line by line. Look for characters that can resemble one another: O and 0, I and 1, or a punctuation mark that disappeared. Check a reference such as AB-101 against the image rather than deciding which version looks more plausible.

Use headings, bold, italic, underline and lists when they help the new document. Formatting edits do not repair missing content automatically. If a scan contains two columns, a table or handwritten notes, reconstruct the reading order yourself and compare it against the page.

The batch button skips pages that have been recognized or manually edited. If a page must be recognized again, its separate button explicitly replaces that page's text. Cancellation keeps finished pages and manual corrections; failed pages can be retried.

Pick an export for the next task

DOCX is useful when a recipient needs to continue editing paragraphs. Headings and emphasis are retained, and list markers remain editable text. This version does not create advanced Word numbering definitions, tables or a reconstruction of the scanned geometry.

TXT is useful for copying text into another system. It discards rich formatting and separates source pages with a form-feed character. A new PDF provides selectable edited text on A4 pages; each source page starts on a new output page and long text wraps onto additional pages.

Text PDF supports Latin, Cyrillic and Greek. For other writing systems or emoji, export corrections as DOCX or TXT. The separate PDF with scan and search format keeps the page image and recognized OCR text.

Keep the scan image and add search

Choose PDF with scan and search and recognize every selected page first. An invisible layer makes recognized words searchable and copyable. Formatting text in the editor does not change the scan image. Use zoom and preview rotation to inspect the original, or switch between original and text on a phone.

Changing the words manually disables this format to prevent exporting a stale hidden layer. Export corrections through Word, TXT or text PDF. Recognizing that page again enables searchable-scan export but replaces its corrected text. Bold or italic styling alone does not change the words or disable export.

Source PDF pages are rasterized at up to 150 dpi, reduced for large pages to stay within pixel limits. PDF page sizes and selected order are retained. This is a new file: original digital signatures, forms, links and bookmarks are not carried over. Tables remain images, and copied reading order depends on OCR and the reader. Verify important lines by searching and copying them.

Searchable PDF currently supports left-to-right text. For Arabic, Hebrew and other right-to-left scripts, choose Word or TXT: the OCR engine can reverse their character order in a PDF text layer.

When recognition is unreliable

Low contrast, skew, tiny characters, glare, handwriting and complex layouts can reduce OCR quality. Start with a clearly printed page and check the result before processing a large selection. If a source is a digital PDF with selectable words, try Split PDF's TXT extraction rather than recognizing a rendered image.

Use the original as the reference for every important field. The editor is a correction tool, not evidence that a guessed number or name is accurate. FileFlow does not certify recognition accuracy or document authenticity.

Finish the session by downloading

Edits live only in the current tab. Refreshing or leaving the page discards the document state. Create the output, download it and open it before closing the editor. Keep the original scans separately.

Input is capped at 50 files and 100 MiB combined. PDF/image sources can contain at most 500 pages combined, but only 20 are selected for one export. Each text page allows up to 20,000 characters, with 200,000 per export. Device memory can impose a lower practical limit.

Worked example: assemble two scanned letters

Open two page images. Choose their printed language and recognize them. Suppose the first letter contains reference AB-101 while OCR returns AB-IOI. Replace the guessed characters only after checking the original. Make the subject line a heading and review the second letter the same way.

Move the second page above the first if that is the required reading order. Export DOCX, open it and verify both letters, their order and the corrected reference. If you also need a stable text copy, export TXT from the same edited pages. This example illustrates correction; it is not a measured OCR accuracy claim.

Practice file

A FileFlow-created raster-only PDF. Choose English recognition, correct the result and compare DOCX with the original pages.

Download a fictional two-page English scan for practice

Try it in FileFlow

  1. Open PDF scans or images, choose the recognition language and select up to 20 pages.
  2. Recognize unfinished pages, review each original and correct its text. Reorder or exclude pages as needed.
  3. Choose Word, TXT or PDF, create the document and download it before leaving this tab.

For two scanned letters, add both files, choose their recognition language and select the required pages. Run recognition, compare each original with its text and correct names and dates. Export DOCX for further editing. If you choose PDF with scan and search, it uses the OCR layer rather than your manual text edits; rerun recognition after changing the source and check search in a PDF reader.

Before you finish

  • Check every included page, even if the OCR text looks fluent.
  • Verify identifiers, dates, names and amounts against the image.
  • Review page order and exclude unneeded pages.
  • Download and open the final document before refreshing the tab.

Limits and privacy

Up to 50 files and 100 MiB combined, with up to 500 source PDF pages; select at most 20 pages. Images: JPG, PNG or WebP, up to 20 MiB and 8 million pixels each. Up to 20,000 characters per page and 200,000 per text export. Text PDF supports Latin, Cyrillic and Greek. PDF with scan and search uses OCR language support, with up to 20 MiB per recognized page and 100 MiB combined. Language models require a network connection. Editable tables and columns are not reconstructed. PDF with scan and search currently supports left-to-right text. For Arabic, Hebrew or other right-to-left text, export Word or TXT.

Processing runs in your browser. File contents, names and recognized text are not sent to FileFlow servers or analytics. Loading the site and OCR models still uses the network. Keep your original until you have checked the download.

This information page does not accept files. Open the tool to process your document in your browser.

Frequently asked questions

Does the export preserve the original scanned layout?

Word and text PDF create a new document from the edited text. TXT keeps plain text only. Choose PDF with scan and search to keep the scan image with an invisible OCR text layer. It uses recognized text, not manual text changes. Source PDF pages are rasterized at up to 150 dpi; page sizes and selected order are retained. Tables and columns are not rebuilt as editable objects. Check OCR before downloading. PDF with scan and search currently supports left-to-right text. For Arabic, Hebrew or other right-to-left text, export Word or TXT. If export is blocked after rotating or replacing a source, recognize that page again: stale OCR cannot be used for searchable PDF. For right-to-left OCR use DOCX or TXT. Download before closing the tab; this workflow has no server backup.

Are my files uploaded?

Processing runs in your browser. File contents, names and recognized text are not sent to FileFlow servers or analytics. Loading the site and OCR models still uses the network. Keep your original until you have checked the download.

Technical references

Related reading