Get High-Quality OCR with Ollama-OCR in Just a Few Lines of Code

Llama 3.2-Vision is a multimodal large language model available in 11B and 90B sizes, capable of processing text and image inputs and generating text output. It excels at visual recognition, image reasoning, captioning, and answering image‑related questions, outperforming many open‑source and closed models on industry benchmarks.

This article introduces the open‑source ollama-ocr tool, which by default uses a locally running Llama 3.2-Vision model to accurately recognize text in images while preserving the original format.

https://github.com/bytefer/ollama-ocr

Features of Ollama-OCR

High‑precision text recognition with Llama 3.2-Vision, retaining original layout.

Supports multiple image formats: JPG, JPEG, PNG.

Customizable recognition prompts and model selection.

Optional Markdown output format.

Llama 3.2-Vision Application Scenarios

Handwritten Text Recognition

OCR Recognition

Image Question Answering

Environment Setup

Install Ollama

Before using Llama 3.2-Vision, install Ollama, a platform for running multimodal models locally.

Download Ollama from the official website for your operating system.

Run the installer and follow the prompts.

Install Llama 3.2-Vision 11B

After Ollama is installed, run the following command:

ollama run llama3.2-vision

Install Ollama-OCR

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

Using 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);
}

The test image is shown below:

The resulting output is:

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.

Output Markdown

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);
}

Test image:

The resulting Markdown output is displayed in the following screenshot:

If you prefer to use an online Llama 3.2-Vision model, you can try the llama-ocr library.

References

ollama-ocr: https://github.com/bytefer/ollama-ocr

Ollama: https://ollama.com/

Llama 3.2-Vision 11B: https://ollama.com/blog/llama3.2-vision

llama-ocr: https://github.com/Nutlope/llama-ocr