Use vision AI to extract text from scanned PDFs and photographed pages, preserving layout — including foreign-language scripts, where ordinary OCR usually fails.
Pro AI tools send document content to xAI.
AI Scan to Text is a Pro feature — included in PDFwarp Pro at $9 USD/month along with every other tool and unlimited conversions.
This is not Tesseract and it does not stamp a hidden text layer onto your scan. Poppler renders the first 10 pages at 200 DPI to JPEG; each page image is sent to xAI vision; the returned text is typeset into a new PDF (or Word for non-Latin scripts). The original page pictures are not in the download. Pro only.
Split a long scan, OCR each chunk, then Merge the text PDFs (or concatenate the Word files). The original scans stay separate if you still need the pictures.
Neat block print often works. Cursive and rushed handwriting often do not. If the output is garbled, the source was the problem.
PDF → Word looks for a text layer and will refuse an image-only file. This tool is the handoff that creates words the free converters can read.
convert_from_path(..., dpi=200, fmt=jpeg), then each of the first 10 images is JPEG-encoded at quality 85 and sent to xAI vision. Pages that error are skipped (logged by type only); if every page fails, the route errors.
build_pdf_output writes 'Extracted Text' plus paragraphs. There is no attempt to place words on the original image. That is why CJK/Arabic become a .docx (Noto via Word) instead of tofu in ReportLab.
A 10-page run is 10 vision requests inside one user-facing operation, and still counts as one increment_usage for the daily cap. Very dense pages can still hit per-request token limits — split if a page comes back empty.
Here it means: rasterize the page, ask a vision model to type what it sees, write a new document. It is not a text layer merged back into the original PDF.
Clean printed pages are usually strong. Cursive, stamps, and very low-resolution phone photos are weaker. There is no published accuracy percentage we can stand behind.
10 per upload. Split, run, merge the text outputs if you need more.
No. You get a new document containing the extracted words. The photos of the pages stay in the file you already have.
AI OCR requires a Pro account. Upgrade at pdfwarp.com/pricing.