Translate scanned PDFs. Even the ones nobody else can read.
Drop a scan, a photo, or an image-only PDF. Fily runs dual-backend OCR (vision-language model + automatic fallback), translates the text, and gives you back a searchable PDF or Word file with the layout preserved.
What is Scanned PDF?
A scanned PDF is a PDF whose pages are images rather than selectable text. Common sources: documents photographed with a phone, documents scanned with a flatbed or sheet-fed scanner, faxes saved as PDF, image-based exports from legacy systems, old contracts, medical records, ID documents, blueprints.
You can tell a PDF is scanned because you can't select the text — clicking and dragging just highlights image regions. Standard translation tools fail silently on these: they see zero translatable content and return an empty file. Native PDFs (PDFs with selectable text generated from Word, InDesign, or LaTeX) are a different problem with a different pipeline.
Why Scanned PDF is tricky for AI translation
- OCR accuracy varies wildly. A clean letter-size scan at 300 DPI is easy. A faxed contract at 150 DPI with a coffee stain is brutal. Most tools use a single OCR engine and silently produce garbage when it fails.
- Layout reconstruction: once text is extracted, putting it back where it came from — paragraphs, columns, tables, headers — is harder than the OCR itself.
- Multilingual documents: a medical record may have Spanish patient notes inside an English form. Single-language OCR drops segments it doesn't recognize.
- Tables and Gantt charts: numeric grids and structured layouts confuse most OCR pipelines into reading rows as flowing text.
- Handwriting and signatures: hard to distinguish from typed text. Hard to translate. Hard to preserve.
- Stamps and headers in non-Latin scripts: Cyrillic, Arabic, CJK headers on otherwise-Latin documents need separate recognition.
How Fily handles Scanned PDF
- Dual-backend OCR: primary backend runs a frontier vision-language OCR. On JSON-decode errors or truncated output, the pipeline automatically fails over to an alternate backend — different transport, more resilient on long outputs.
- Per-page processing: each page is OCR'd independently, then assembled. A failure on one page doesn't kill the whole document.
- Block-level extraction: OCR returns structured blocks (paragraphs, tables, headers) with positions, not flat text. Layout is reconstructed from blocks, not guessed from line breaks.
- Translation per block: each block is translated as a unit with surrounding context. Tables get table-aware translation that preserves columns.
- OCR-only mode: if you only want the text out (no translation), select "OCR only" at upload. Fily skips translation and delivers a searchable DOCX.
- Output options: searchable PDF (text layer on top of original images) or DOCX with reconstructed layout. 12-step QA runs on both.
Pipeline: pdf_qa_12step_v2@2.0.0 (translation) · image_qa_12step_v2@1.0.0 (OCR-only)
The Scanned PDF workflow with Fily
Upload
Drop your .pdf (single or batch ZIP). Optional: glossary, TM, style guide.
Process
Fily runs the Scanned PDF pipeline + 12 QA steps. Typical job: 10–20 minutes.
Download
Same format, ready to deliver. QA report HTML attached.
Common use cases we see daily: medical records for healthcare LSPs, blueprints and Gantt charts for engineering firms, immigration documents for legal teams, school forms for public education districts, and historical archives for cultural institutions.
Frequently asked about Scanned PDF
Other formats Fily translates
Ready to translate a Scanned PDF file?
No card. No setup. Upload one file and see the output.