Scan PDF to Markdown with local OCR
PDFLLM converts scanned PDFs to Markdown with local PP-OCRv6 and document-aware post-processing. It renders the page, applies conservative grayscale and contrast preprocessing, recognizes English and Chinese text, and reconstructs reading order from the resulting coordinates. Instead of hiding OCR uncertainty, PDFLLM keeps alternate candidates, confidence signals and source regions available for review before you export the document.
Product information updated September 2, 2026What improves OCR reliability?
Evidence-aware routing
Pages with reliable embedded text avoid unnecessary OCR; image-only or corrupted text regions are routed through recognition.
Conservative preprocessing
Grayscale, contrast and scale adjustments make faint scans easier to read without aggressively rewriting the original evidence.
Human-verifiable output
Low confidence, garbled text, empty scans and alternate OCR candidates remain visible beside the source page.
How are scanned PDFs recognized?
Render at useful resolution
Each scanned page is converted to an image at a resolution suited to local recognition, including tiled handling for unusually tall pages.
Recognize text and coordinates
PP-OCRv6 returns text regions and positions. The pipeline groups them into lines, blocks, columns and recoverable tables.
Score document risk
A multi-dimensional trust report highlights recognition, structure, source coverage and completeness risks before AI use.
Local OCR versus cloud OCR
The right option depends on document sensitivity and layout difficulty. PDFLLM optimizes for privacy, low marginal cost and verifiable output on common documents.
| Trade-off | Cloud OCR | PDFLLM local OCR |
|---|---|---|
| Document transfer | Usually uploaded | Stays in the browser by default |
| Per-page cost | Usage-based | No OCR API charge |
| Cold start | Server dependent | Model downloads on first use |
| Difficult layouts | Varies by provider | Visible flags and source review |
Common scanned-PDF questions
Which OCR languages are supported?
The current launch configuration focuses on English and Simplified Chinese. Mixed pages are supported, while unusual scripts may require a future language pack.
Why can OCR still be wrong?
Blur, skew, handwriting, decorative fonts, stamps, equations and compressed scans can remove visual evidence. PDFLLM reports uncertainty because no OCR system can recover information that is not present.
Is local OCR free?
The OCR engine runs locally without a per-page API bill. Account plan limits still apply to successful conversions performed through the hosted product.