Parse long documents and PDFs with open OCR

Unlimited OCR is built for document parsing across long screenshots, scanned pages, and multi-page PDFs. Instead of treating OCR as a one-line text extraction task, it aims to preserve more of the document structure over longer inputs.

That matters when the source is not a neat receipt or a single screenshot. For example, you can convert a PDF into page images, run the model across those pages, and inspect whether it keeps tables, sections, and long text blocks coherent enough to use downstream.

Start with the official Hugging Face Space and upload a real document page, not a perfect demo image. A good first test is a PDF page with headings, tables, and dense text. If it handles that reasonably, try a longer multi-page document locally.

For local use, the model card includes Transformers examples, a vLLM recipe, and an SGLang path. The local workflow is reproducible from public weights and code, but it is still a GPU workflow and uses `trust_remote_code`, so treat the browser demo as the quick first filter.

This is not a lightweight CPU OCR utility. Local use expects a modern GPU stack, pinned runtime versions, and custom model code. For many readers, the Space is the right first test before investing in setup.

• Extract text from long screenshots, scanned forms, and PDF pages.
• Convert multi-page documents into text for search, cleanup, or summarization.
• Test whether table-heavy pages survive OCR well enough for downstream processing.
• Compare an open OCR workflow against a closed document parsing API.
• Build a local document extraction path when privacy or repeatability matters.

Try the demo

View model page