It all started with my little brother. He asked me something from his textbook, and like any busy older sibling, I gave him the universal “wait a minute.” That “minute” never came. But the idea stuck — what if text could just... come to life?

And just like that, inspired by the magical newspapers from Harry Potter (yes, the ones that move), Artikle was born. It's not wizardry — it's code. But to non-techies, the results are just as magical.

Artikle is your mini creative studio in a Python pipeline. Give it an image of text—be it a textbook page, a news clipping, or even that 3rd grade storybook—and it’ll:

• 📸 Read the text using OCR (Tesseract).
• 🧠 Summarize the key idea using your favorite LLM (Gemini, OpenAI, or Ollama).
• ✏️ Craft an image generation prompt.
• 🎨 Generate a stunning image using DALL·E or a local diffusion model.

Yes, all of that. From one picture. You’re welcome.

• 🖋️ Image-to-Text Extraction
  Snatch text straight out of images (like textbooks, articles, or even that weird meme) using Tesseract OCR.

• 📚 Smart Summarization
  Summon the power of LLMs—ChatGPT, Gemini, or Ollama—to compress walls of text into crystal-clear summaries.

• 🎯 Built-in Prompt Engineering
  Automatically converts summaries into art-ready prompts, so you don’t have to channel your inner poet.

• 🖼️ Stunning Image Generation
  Turn text into eye-candy using OpenAI’s DALL·E or locally hosted diffusion models like Stable Diffusion XL or Waifu Diffusion.

• 🔄 Model-Switching Magic
  No need to crack open the code! Just tweak a .env file to swap between LLMs or image generators.

• 💡 Prompt Customization? EZ.
  Want to fine-tune the image prompt style? Just edit a file in the prompts directory —no digging through code required.

• 🌍 On-Prem Ready
  Run everything locally. No cloud? No problem. Perfect for secure, offline, or private deployments.

• 🔌 API-First Architecture
  Use Artikle as a service! Easily integrate it into other apps, bots, or tools without rewriting logic.

• 🧪 Minimal Dependencies, Max Impact
  Clean codebase with modular components—hackable, forkable, remixable.

• ✨ Harry-Potter-Vibes? Definitely.
  Because why shouldn’t textbook learning feel like flipping through a magical newspaper?

git clone https://github.com/AsutoshPati/artikle.git
cd artikle

[Optional, but highly recomended] Create a virtual environment to keep things clean, isolated, and drama-free (avoid any library conflicts):

python3 -m venv venv
source venv/bin/activate  
# On Windows use `venv\Scripts\activate`

Now install the required packages

pip install -r requirements.txt

• Linux

  sudo apt install tesseract-ocr

• macOS

  brew install tesseract

• Windows
  Download the installer from Tesseract UB Mannheim
  And add the installation path to your system’s environment variables.

🔗 Tesseract Documentation

Ollama runs models locally on your machine, like LLaMA or Mistral.

• Install from: https://ollama.com/download
• Run locally with:

  ollama run <model-name>
  # example: ollama run llama3.2:3b

• 🔗 Docs: Ollama Dev Guide

• Sign up: https://auth.openai.com/create-account
• Go to API keys → generate one.
• Docs: OpenAI API Guide

• Visit: https://makersuite.google.com/app
• Create a project, get your API key.
• Docs: Gemini API Reference

Copy from the template and make it your own:

cp sample.env .env

Then, update like so:

# Server Config
HOST="0.0.0.0"
PORT=9090

# LLM Model Config
USE_MODEL="OLLAMA"
#### Options
#### GEMINI - API key needed
#### OLLAMA - Local (Prior installation required)
#### OPENAI - API key needed

# Image Gen Model Config
IMG_GEN_SERVICE="DIFFUSION"
#### Options
#### OPENAI - API key needed
#### DIFFUSION - Local (Prior installation required)

## OpenAI Config
OPENAI_API_KEY=<your_openai_api_key>
OPENAI_LLM_MODEL="gpt-4o-mini"
OPENAI_IMG_MODEL="dall-e-3"

## Gemini Config
GEMINI_API_KEY=<your_gemini_api_key>
GEMINI_MODEL="gemini-2.0-flash"

## Ollama Config
OLLAMA_MODEL="llama3.2:3b"

## HuggingFace Credential
HF_TOKEN=<your_huggingface_token>

## Diffusion Model
DIFFUSION_MODEL_OPT=2
#### Options
#### 1 - Stable Diffusion by RunwayML
#### 2 - Stable Diffusion XL
#### 3 - Flux v1 Dev (Do not use; requires very high GPU)
#### 4 - Waifu Diffusion
#### 5 - DeepFloyd IF (Do not use; requires very high GPU)

Once everything is set up; You’re ready to turn boring old text into magic!

python main.py

Now sit back and let Artikle turn words into worlds.

Endpoint: /upload
Method: POST
Content-Type: multipart/form-data

Payload Example:

style: realistic  # options: cartoon, comic, realistic
file: your_image.png

Sample Response:

{
  "summary": "Photosynthesis is the process...",
  "image_prompt": "A lush green forest under sunlight showing energy transfer",
  "generated_image": "http://localhost:9090/images/generated.png"
}

💡 You can also use the prompt in your own image tools (ChatGPT web UI does wonders with it).

Swagger Endpoint: /docs

• Click “Try it out”
• Choose your desired image style
• Upload the image with the text content you want visualized
• Hit "Execute" and relax while the magic happens

Change these in your .env, no code edits needed:

USE_MODEL="GEMINI"  # or OPENAI or OLLAMA
IMG_GEN_SERVICE="DIFFUSION"  # or OPENAI
DIFFUSION_MODEL_OPT=1  # 1 to 5 options available

Diffusion Options:

😞 Not happy with the response? Want to upgrade the prompt?
🤩 It’s EZ!

You don’t even need to touch the code — just head over to the prompts directory and tweak the templates to your heart’s content. Your creativity, your rules.

By ChatGPT

• 🔊 Add Text-to-Speech narration
• 🎮 Let users fine-tune image styles
• 🧠 Support more LLM/image models
• 🌐 Host a web UI for non-tech users

You made it to the bottom? You're my kind of developer. Thanks. ❤️
This project was written not just to solve a problem, but to spark a little joy and curiosity.
Feel free to fork, hack, and build something even cooler with it. And, don’t forget to ⭐ give a star if you liked the project — it really helps and keeps the magic going!

This project is licensed under the MIT License - see the LICENSE file for details.

• Inspired by my curious little brother 🧒
• And Harry Potter, for making newspapers awesome again.
• Thanks to the incredible models by OpenAI, Hugging Face, Google AI, and Ollama.