⚡️ Build quick LLM pipelines for all your image specific tasks

Picachain is a framework for building complex non language pipelines for your LLM application. Build pipelines for your images, charts and graphs, parse, extract and generate using opensource models with ease, thanks to support for huggingface and diffusion libraries.

🌉 Things you can build with Picachain

• Image Search Engines: Build quick search engines for your images using ChromaDB and Pinecone for vector storage and retrieval.
• Image Generation: Generate images with ease, thanks to support of Stable Diffusion, SDXL and SDXLturbo.
• Chart Parsing: Parse chart/graph from images and extract meaningful data and build a question-answering system on top of this using an LLM.

📌 Install Picachain

pip install picachain

🥇 Example Usage

Build a quick image search engine

Use ChromaDB or Pinecone for storage with CLIP embeddings. Check out a demo

Fig: Image Search Engine Overview

# initiate embedding, datastore and retriever
embedding = ClipEmbedding()
datastore = ChromaStore("test-collection")
retriever = ImageRetriever(datastore, embedding, images)
img_chain = ImageSearchChain.from_image(retriever, embedding, img)
result = img_chain.similar_images(top_k=3)

## Output 
# [(img, score), (img, score)]

Fig: Demo

Conversation with Charts/Graphs

Create a conversation chain and ask questions from chart/graph images.

Fig: Conversation chain for Q&A with charts/graphs

Check out the example notebook under examples directory.

💡 Contributing

As an open-source project, we are open to all kinds of contribution, be it through code, documentation, issues, bugs, or even feature suggestions.

Feel free to check out Contribution guide for more details.

🔧 Dependencies

We use poetry as the package manager. Make sure to refer to pyproject.toml for more details on dependencies.

cd picachain
pip install poetry
poetry install