PaddleOCR-VL

PaddleOCR-VL

Reads document images and returns their text, in a vision-language model under one billion parameters. For people who parse English and Chinese documents with PaddleOCR on their own machine.

Made by PaddlePaddle on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: PaddleOCR-VL is a compact vision-language model from PaddlePaddle that turns document images into text. It has about 959 million parameters and is built on ERNIE-4.5-0.3B-Paddle. Its title describes it as a model for multilingual document parsing.

Who it suits: People who need text pulled out of scans, photos of pages and other document images, and who want it done locally.

Industry | Who uses it | How they use it | Example
Office work | Admin staff | Turn scanned pages into editable text | Digitize a stack of paper forms
Development | Developers | Add document reading to an app through PaddleOCR | Parse uploaded pages in a document tool
Research | Students and researchers | Extract text from papers and book pages | Read a scanned chapter in English or Chinese

If your documents are already digital text, you can skip it.

Where to use it:

• Your own computer with Python, on Windows, macOS or Linux
• Programs built on the PaddleOCR library

Use case | Who | What they get back
Document parsing | Developers | Text read from a page image
Scan digitizing | Office teams | Text from scanned paper
Bilingual documents | Students | Text from English and Chinese pages

How to set it up:

• Python with pip
• About 1.9 GB of disk space for the BF16 weights, or about 479 MB at 4-bit (an estimate)
• The PaddleOCR library, which the model is made for

1. Install the Hugging Face hub tool.
2. Download the model files.
3. Follow the model card for the PaddleOCR steps that load and run them.

pip install -U huggingface_hub
hf download PaddlePaddle/PaddleOCR-VL

How to use it: You give the model an image of a document and it returns the text it reads. The part of the card available here does not show a run command, so use the full model card for the exact call.

Example: You point it at a scanned page from a report. It reads the page and gives you the text back, which you can then search or edit.

For newcomers and experts: If you are new, begin with one clean scan and compare the output with the page by eye. The model is small, so a normal computer can hold it.

For experts, this is a vision-language model of 959M parameters on top of a 0.3B ERNIE base. That size is a trade of reading accuracy for low memory and fast start-up, which suits batch document work.

You load a page image
The vision part reads the layout
The language part writes the text
You get the page text back

Good to know: The Hugging Face record lists English and Chinese as the languages.

• License: Apache 2.0, as the Hugging Face record lists it.
• Cost to run: no API key or account is named for a downloaded copy. You supply the computer.
• Data it reaches: downloading reaches Hugging Face. The card shown here does not describe it reaching your files or the web on its own.
• Not for: the card shown here names no banned uses. It is aimed at reading documents, not at chatting.

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: PaddleOCR
• Parameters: 959M
• Weights: about 1.9 GB in BF16; about 479 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 12,780
• Likes: 1,685
• License: apache-2.0
• Languages: en, zh
• Based on: baidu/ERNIE-4.5-0.3B-Paddle
• Quantized versions: 9
• Fine-tunes: 7
• Tags: paddleocr, safetensors, paddleocr_vl, ernie4.5, paddlepaddle, image-to-text, document-parse, layout, table, formula, chart, image-text-to-text

Links

• Model: https://huggingface.co/PaddlePaddle/PaddleOCR-VL
• Files: https://huggingface.co/PaddlePaddle/PaddleOCR-VL/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download PaddlePaddle/PaddleOCR-VL

Then follow the usage in its model card. Nothing in this kit runs by itself.