What is the difference between the PDF and the Word document?

The PDF is the original scan with a searchable text layer. For printed pages that layer is a faithful Tesseract OCR layer; for handwriting it is a machine transcription overlaid invisibly, and the scan stays the visible, authoritative page. The Word document is a separate, editable reconstruction — printed pages with structure (headings/lists/tables) and handwriting as the verbatim transcription — which you should verify against the scan.

Does it transcribe handwriting, and can I trust it?

Yes — handwritten pages are transcribed verbatim by a vision-language AI. It marks anything it cannot read as [illegible] and uncertain words as [?: best guess], and it does not invent or correct text. A per-page confidence note shows how many tokens are uncertain. Treat the transcription as a careful machine reading to verify against the original, not as ground truth.

Can I scan a document that mixes printed text and handwriting?

Yes. In Auto mode each page is classified as printed or handwritten and routed to the right engine, then everything is assembled into one ordered PDF (and Word document). A notebook with a printed form and handwritten answers is exactly the target.

Can I reorder the pages?

Yes. On the dedicated tool page you can drag pages into the correct reading order before processing; the final PDF and Word document follow that order.

What languages are supported for printed OCR?

Printed OCR uses Tesseract, which supports many languages. The available language packs depend on the server; you can select one or combine several (e.g. English + French) for the printed pages.

Is the searchable PDF a faithful copy of my scan?

Yes for printed pages: OCRmyPDF adds an invisible text layer under the untouched scan and outputs PDF/A. For handwriting pages the original scan is still the visible page; the transcription is an invisible, clearly-labeled machine layer.

Do my files go to a third-party AI service?

No. All processing — OCR and the handwriting AI — runs on our own GPU server. Your uploads are designed to be deleted automatically within 24 hours and no account is required.

AI Document Scan — Scanned Pages & Handwriting to Searchable PDF / Word

Turn scanned documents and handwritten field notes into a searchable, multi-page PDF and an editable Word document — printed pages get a faithful Tesseract text layer, handwriting is transcribed by a vision AI with uncertainty marked, never fabricated.

AI Document Scan turns one or more scanned pages into a searchable, multi-page PDF and, optionally, an editable Word (DOCX) document. It handles two very different kinds of page and is honest about the difference. Printed/typed pages go through deterministic OCR (OCRmyPDF + Tesseract): the PDF is the original scan with an invisible, searchable text layer grafted underneath — the resolution is untouched and nothing is rewritten. Handwritten pages (detective field notes, a notebook, scratch paper, a napkin) are read by a vision-language AI that transcribes the writing verbatim, marking anything it cannot read as [illegible] and uncertain words as [?: best guess] — it never invents or "cleans up" what you wrote. A single job can mix both, page by page, and assemble one ordered output. Everything runs on our own GPU server, never a third-party API.

Loading AI Document Scan…

Supported formats

Input formats

PNG
JPG
JPEG
WEBP
TIFF
BMP
HEIC

Output formats

PDF
DOCX

Add up to many pages per job; each page image has an individual size cap (~25 MB).
A searchable PDF is always produced; the editable Word document is optional.
Printed pages use deterministic OCR (OCRmyPDF + Tesseract) for a faithful, searchable text layer.
Handwriting is transcribed verbatim by a vision AI with [illegible]/[?] markers preserved — it is a machine transcription, not the authoritative page.
The Word document and any AI summary are reconstructions — verify them against the original scan.

How to use it

Add your pages
Drop in one or more scanned images (PNG, JPEG, WebP, TIFF, BMP, HEIC) — printed pages, handwritten notes, or a mix.
Reorder if needed
Drag pages into the right reading order before anything is processed.
Choose mode & outputs
Pick Auto, Printed, or Handwriting; choose a searchable PDF and optionally a Word document, plus language(s) for printed OCR.
Watch the pipeline run
A live checklist tracks classification, reading, verification, and packaging with rising progress.
Review & download
A built-in PDF viewer opens on completion; download the searchable PDF and, if requested, the Word document and a separate AI summary.

What this tool does

Converts a scanned image (or many) into a searchable, multi-page PDF — selectable, copy-pasteable text over the original scan.
Transcribes handwritten notes to text: the PDF keeps the original scan as the visible page and overlays an invisible machine-transcription layer so it is searchable, clearly labeled as a machine transcription.
Optionally produces an editable Word (DOCX) document — printed pages reconstructed with structure (headings, lists, tables), handwriting as the verbatim transcription, each labeled a reconstruction.
Auto mode classifies each page as printed or handwritten and routes it to the right engine, so mixed documents come out as one ordered file.
Lets you drag-reorder pages before processing and pick output language(s) for printed OCR.
Preserves uncertainty: [illegible] and [?: guess] markers survive end to end, and a per-page confidence note is shown — nothing is fabricated.

Why it matters

A photo of a document is not a document: without an OCR text layer you cannot search, select, or quote it. This adds that layer while keeping the original scan as the authoritative page.
Handwritten notes are some of the hardest pages to digitize; a verbatim, uncertainty-marked transcription is far more useful — and more honest — than a confident guess.
The faithful-vs-reconstructed distinction matters for investigative and archival work: the printed PDF is faithful, the DOCX and any summary are reconstructions you should verify against the scan.

Common use cases

Make scanned paperwork searchable

Turn a stack of scanned contracts, forms, or receipts into one searchable PDF you can grep and quote.

Transcribe handwritten field notes

Convert detective notes, lab notebooks, or interview scribbles into editable text with uncertainty preserved, not fabricated.

Image to Word (OCR)

Get an editable .docx from a scanned printed page, with headings, lists, and tables reconstructed.

OCR notebook pages

Digitize mixed notebook pages — a printed form with handwritten answers — into one ordered, searchable document.

Archive and quote old documents

Add a searchable text layer to scanned archival pages without touching the original scanned image.

Advanced details

Printed PDF is faithful: OCRmyPDF grafts an invisible Tesseract text layer under the original scan and outputs PDF/A — the scanned pixels are never altered, only made searchable.
Handwriting PDF is the original scan plus an invisible machine-transcription layer positioned per line from the vision model’s grounding boxes (a full-page invisible block when boxes are unavailable), with a visible "machine transcription — verify against original" stamp.
Handwriting is read by a general vision-language model (qwen3-vl) tuned for verbatim transcription; an optional second-opinion engine (PaddleOCR-VL, or TrOCR as a lighter fallback) independently re-reads the page and disagreements are flagged [?: vlm | other], never silently merged.
An optional verification pass re-reads only the tokens you flagged as uncertain and may resolve them from the image — but it is instructed to leave confident text untouched and never add content that is not visibly written.
DOCX is a reconstruction: printed pages are parsed to structured Markdown (PaddleOCR-VL preferred, Docling fallback) preserving tables and headings; handwriting sections carry the verbatim transcription with a bold "verify against the original scan" note; Pandoc renders the final .docx.
The optional AI summary is written to a separate, clearly-labeled file ("AI-generated summary — not a verbatim transcription") and never modifies the verbatim outputs.
Capabilities degrade gracefully: if the handwriting engine is offline, printed OCR still works; if no structure engine is reachable, DOCX is disabled but the PDF still ships.

Why Media Manipulator

Honest by design: the printed PDF is faithful (original scan + searchable layer), while the Word document and any AI summary are labeled reconstructions you should verify against the scan.
Verbatim, never fabricated: handwriting is transcribed exactly, with [illegible] and [?: guess] markers preserved and surfaced as a per-page confidence note — and an optional independent second opinion flags disagreements instead of hiding them.
Runs entirely on our own GPU server: your scans and notes never go to a third-party AI provider, and no login is required.

Privacy & file handling

Your uploads are processed on our own servers and are designed to be automatically deleted within 24 hours. We do not share your files with third-party AI providers, and AI features run on a local GPU server we operate. No login or account is required.

Frequently asked questions

What is the difference between the PDF and the Word document?
The PDF is the original scan with a searchable text layer. For printed pages that layer is a faithful Tesseract OCR layer; for handwriting it is a machine transcription overlaid invisibly, and the scan stays the visible, authoritative page. The Word document is a separate, editable reconstruction — printed pages with structure (headings/lists/tables) and handwriting as the verbatim transcription — which you should verify against the scan.
Does it transcribe handwriting, and can I trust it?
Yes — handwritten pages are transcribed verbatim by a vision-language AI. It marks anything it cannot read as [illegible] and uncertain words as [?: best guess], and it does not invent or correct text. A per-page confidence note shows how many tokens are uncertain. Treat the transcription as a careful machine reading to verify against the original, not as ground truth.
Can I scan a document that mixes printed text and handwriting?
Yes. In Auto mode each page is classified as printed or handwritten and routed to the right engine, then everything is assembled into one ordered PDF (and Word document). A notebook with a printed form and handwritten answers is exactly the target.
Can I reorder the pages?
Yes. On the dedicated tool page you can drag pages into the correct reading order before processing; the final PDF and Word document follow that order.
What languages are supported for printed OCR?
Printed OCR uses Tesseract, which supports many languages. The available language packs depend on the server; you can select one or combine several (e.g. English + French) for the printed pages.
Is the searchable PDF a faithful copy of my scan?
Yes for printed pages: OCRmyPDF adds an invisible text layer under the untouched scan and outputs PDF/A. For handwriting pages the original scan is still the visible page; the transcription is an invisible, clearly-labeled machine layer.
Do my files go to a third-party AI service?
No. All processing — OCR and the handwriting AI — runs on our own GPU server. Your uploads are designed to be deleted automatically within 24 hours and no account is required.