Embedding Galaxy

A 3D image explorer that uses Google Gemini's embedding model to map images into semantic space. Upload images and watch them arrange themselves in a galaxy based on visual and semantic similarity. Similar images cluster together, connected by edges.

Setup

1. Clone the repo and install dependencies:

   uv sync

2. Create a .env file with your Google API key:

   GOOGLE_API_KEY=your_key_here

   Works great with free rate limits from Google AI Studio.

3. Run the server:

   uv run python main.py

4. Open http://localhost:8000

Usage

• Upload: drag and drop images onto the page, or use the upload button (PNG/JPEG)
• Navigate: drag to look around, scroll to fly forward/back, WASD + Space/Shift to move
• Inspect: click any image node to see its similarity connections
• Tune: adjust the connection threshold and node size sliders in the sidebar

How it works

Each uploaded image is embedded using gemini-embedding-2-preview (3072 dimensions). On every page load the embeddings are PCA-reduced to 3D coordinates and cosine similarity is computed between all pairs. Images with similarity above 0.65 are connected by lines.