📘 Free LLM-Based OCR Using OpenRouter (No Cost, Image → Text Extraction)

This project performs OCR (Optical Character Recognition) using free vision-enabled LLMs available on OpenRouter. No paid API keys, no subscriptions — just fully free OCR using models like:

• amazon/nova-2-lite-v1:free
• Any OpenRouter vision-capable free model

This allows you to extract text from images using modern LLMs at zero cost.

---

🚀 Features

• 100% free OCR using OpenRouter’s free-tier models

• Works with:

  • 🌐 Image URLs
  • 🖼️ Local images (converted to Base64)

• Model-agnostic (works with any vision LLM on OpenRouter)

• Preserves reading order and all visible text

• Saves extracted text to a .txt file

• No OpenAI API key needed — only OpenRouter free API

---

🛠 Requirements

• Python 3.7+
• requests
• python-dotenv (optional, for managing API key)

Install:

pip install requests python-dotenv

Add your OpenRouter API key to a .env file:

OPENROUTER_API_KEY=your_key_here

(You can use the free key available on your OpenRouter dashboard.)

---

📘 How It Works

OpenRouter allows sending image URLs or base64 images to free LLMs. This script sends:

• The instruction prompt
• The image as a URL or base64 data URL

And returns clean text extracted by the model.

---

📂 Functions

✔️ encode_image_to_base64(image_path)

Loads a local image and converts it to Base64.

✔️ image_to_text_from_url(image_url)

Downloads the image → sends to LLM → extracts text.

✔️ image_to_text_from_base64(image_base64)

Sends Base64 directly in a data:image/jpeg;base64,... format.

Both functions call a free OpenRouter LLM.

---

🧪 Example Usage

Extract text from a local image

image_path = "image.jpg"
image_base64 = encode_image_to_base64(image_path)
text = image_to_text_from_base64(image_base64)
print(text)

Extract text from an online URL

image_url = "https://example.com/sample.jpg"
text = image_to_text_from_url(image_url)
print(text)

Save the OCR output

with open("extracted_text.txt", "w", encoding="utf-8") as f:
    f.write(text)

---

🔄 Change the Model Easily

Modify:

"model": "amazon/nova-2-lite-v1:free"

To any OpenRouter free vision model (if available):

"model": "qwen2.5-vision:free"
"model": "llama-vision-free"
"model": "openai/gpt-4o-mini:free"

No other changes required.

---

🧠 Why LLM-Based OCR?

Compared to traditional OCR tools, LLM OCR:

• Reads small & faint text
• Preserves formatting
• Handles mixed fonts
• Works on stylized documents
• Understands layout and context

And with OpenRouter free models — you pay nothing.

---

🎯 Use Cases

• OCR for receipts, invoices, IDs
• Text extraction from PDFs (after converting to images)
• Digitizing notes
• Document preprocessing for ML pipelines
• Free OCR for students & hobbyists

---

🤝 Contributing

Feel free to submit pull requests for improvements:

• Model selector menu
• Export to JSON/Markdown
• Batch processing
• CLI tool

---

📄 License

MIT License.