Ollama-OCR for High-Precision OCR with Ollama

Llama 3.2-Vision is a multimodal large language model available in 11B and 90B sizes, capable of processing both text and image inputs to generate text outputs. The model excels in visual recognition, image reasoning, image description, and answering image-related questions, outperforming existing open-source and closed-source multimodal models across multiple industry benchmarks.

Find Your Best Next.js Starters

Find Awesome Shadcn

Llama 3.2-Vision Examples

Handwriting

Optical Character Recognition (OCR)

In this article I will describe how to call the Llama 3.2-Vision 11B modeling service run by Ollama and implement image text recognition (OCR) functionality using Ollama-OCR.

Features of Ollama-OCR

🚀 High accuracy text recognition using Llama 3.2-Vision model
📝 Preserves original text formatting and structure
🖼️ Supports multiple image formats: JPG, JPEG, PNG
⚡️ Customizable recognition prompts and models
🔍 Markdown output format option
💪 Robust error handling

MacOS Vision OCR: Accurate and Fast OCR Tool for macOS

Installing Ollama

Before you can start using Llama 3.2-Vision, you need to install Ollama, a platform that supports running multimodal models locally. Follow the steps below to install it:

1. Download Ollama: Visit the official Ollama website to download the installation package for your operating system.
2. Install Ollama: Follow the prompts to complete the installation according to the downloaded installation package.

Install Llama 3.2-Vision 11B

After installing Ollama, you can install the Llama 3.2-Vision 11B model with the following command:

ollama run llama3.2-vision

How to use Ollama-OCR

npm install ollama-ocr
# or using pnpm
pnpm add ollama-ocr

OCR

import { ollamaOCR, DEFAULT_OCR_SYSTEM_PROMPT } from "ollama-ocr";

async function runOCR() {
  const text = await ollamaOCR({
    filePath: "./handwriting.jpg",
    systemPrompt: DEFAULT_OCR_SYSTEM_PROMPT,
  });
  console.log(text);
}

Input Image:

Output：
The Llama 3.2-Vision collection of multimodal large language models (LLMs) is a collection of instruction-tuned image reasoning generative models in 118 and 908 sizes (text + images in / text out). The Llama 3.2-Vision instruction-tuned models are optimized for visual recognition, image reasoning, captioning, and answering general questions about an image. The models outperform many of the available open source and closed multimodal models on common industry benchmarks.

Markdown Output

import { ollamaOCR, DEFAULT_MARKDOWN_SYSTEM_PROMPT } from "ollama-ocr";

async function runOCR() {
  const text = await ollamaOCR({
    filePath: "./trader-joes-receipt.jpg",
    systemPrompt: DEFAULT_MARKDOWN_SYSTEM_PROMPT,
  });
  console.log(text);
}

Input Image：

Output:

Use MiniCPM-V 2.6 Vision Model

async function runOCR() {
  const text = await ollamaOCR({
    model: "minicpm-v",
    filePath: "./handwriting.jpg.jpg",
    systemPrompt: DEFAULT_OCR_SYSTEM_PROMPT,
  });
  console.log(text);
}

ollama-ocr is using a local vision model, if you want to use the online Llama 3.2-Vision model, try the llama-ocr library.