Unlimited OCR Works 2 Studio

Unlimited OCR Works 2 Studio

Parses uploaded page images or a PDF of up to 20 pages into laid-out Markdown, with live streaming and a rendered view, using the Unlimited-OCR-2 model. It suits people who digitize documents and test OCR.

Made by baidu on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space is a studio for the Unlimited-OCR-2 document parsing model from Baidu. You give it page images or a PDF with at most 20 pages. The raw result appears as it is produced, and a rendered view then lays out each block by the kind it was detected as. It runs on ZeroGPU at Hugging Face and is meant for small-scale testing.

Who it suits: It suits people who turn scanned pages or PDFs into Markdown and want to see how the model handles layout, figures and multi-page documents.

Industry | Who uses it | How they use it | Example
Document processing | Operations and records teams | Turn a scanned report into structured text | Upload a PDF and copy the Markdown
Research | Researchers | Parse papers with figures and page breaks | Check that a block split across two pages comes back as one
AI engineering | ML engineers | Test the OCR model before deploying it | Compare detail modes on the same page

You can skip it if you need to process long documents at scale, since the README sends those to a different serving stack.

Where to use it: You can use it in these places:

• A browser: open the Space's page on Hugging Face and nothing needs installing.
• Windows, macOS or Linux: run your own copy with Python 3.12.

Use case | Who | What they get back
PDF to Markdown | Document teams | A rendered view and Markdown you can copy or download
Multi-page parsing | Researchers | One parse of the whole document, with blocks merged across page breaks
OCR testing | Engineers | Output from three detail modes to compare

How to set it up: You need these first for a local copy:

• Python 3.12 and pip
• A GPU able to run the model, and room for about 7 GB of weights on first start

The online page needs none of this. For your own copy, follow the steps below.

1. Install the Hugging Face tools.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio at the version the Space uses, then the other packages.
5. Start the app.

pip install -U huggingface_hub
hf download baidu/Unlimited-OCR-2 --repo-type space --local-dir Unlimited-OCR-2
cd Unlimited-OCR-2
pip install gradio==6.30.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space in a browser, or start your own copy with the command below. Upload page images or a PDF, watch the parse stream in, then read the rendered view and copy or download the Markdown.

python app.py

Tool | What it does
auto | Tiles a single page in high detail and uses one global view per page for documents
gundam | Forces tiling on every page
base | Forces one global view per page
History strip | Keeps your recent parses within reach

Example: You upload a PDF of several pages. The page streams the raw parse live, then shows a rendered view with figures cropped from the page images and blocks that crossed a page break joined back together.

For newcomers and experts: If you are new to this, upload one page image and leave the detail mode on auto. Compare the rendered view with your original page.

If you know the field, the Space uses Transformers to run the model instead of the SGLang serving stack, which means unlimited-horizon decoding is not available here. A multi-page document goes in one request so the model sees context across pages. Generation follows the model card's setup, with greedy decoding, a guard on page count and a larger repeat-blocking window when a document has several pages.

Upload pages or a PDF
One request parses all pages
The raw parse streams in live
Blocks are laid out by type
You copy or download Markdown

Good to know: A first start fetches roughly 7 GB of model weights. The first parse after booting also waits while the model loads once. The model's tested library versions are installed at runtime inside app.py.

• License: mit
• Cost to run: ZeroGPU time at Hugging Face, and no paid service is named
• Data it reaches: the page images or PDF you upload, and model files downloaded from Hugging Face
• Not for: long documents at scale, which the README says to run with SGLang instead

On Hugging Face, 13 November 2026

• Likes: 17
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/baidu/Unlimited-OCR-2
• Files: https://huggingface.co/spaces/baidu/Unlimited-OCR-2/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.