A 3D image exploration application that maps uploaded images into semantic space using Google Gemini’s multimodal embedding model. Images arrange themselves dynamically into a visual galaxy where similar images cluster together and connect via similarity edges.
Key Features
- Multimodal Embeddings: Embeds images using
gemini-embedding-2-preview(3072 dimensions) to capture both visual and semantic meaning. - 3D Interactive Space: Navigate through an interactive 3D galaxy using mouse and keyboard controls (fly, rotate, inspect).
- Dynamic Clustering & Graph Connections: Visualizes cosine similarity thresholds as graph edges between related image nodes.
- Interactive Controls: Adjust connection threshold cutoffs and node sizes in real time from the sidebar.
- Image Inspector: Click any image node to view metadata and inspect its nearest neighbors.
Setup and Running
# 1. Clone repository and install dependencies with uv
uv sync
# 2. Configure Google AI Studio API key
echo "GOOGLE_API_KEY=your_key_here" > .env
# 3. Start local server
uv run python main.pyThen open http://localhost:8000 to upload and explore images in 3D.