This project is a backend service for generating 3D scenes and videos from text descriptions. It leverages two main tools:
- Three.js Tool: Generates 3D scenes using Three.js
- Cosmos Tool: Generates 3D scenes using NVIDIA's Cosmos API
The project uses a graph-based workflow system with nodes for different processing steps. It leverages Language Models (LLMs) like NVIDIA and Claude for enhancing prompts before generating content.
- LangGraph: Powers the workflow orchestration with directed graphs and state management
- Uses
StateGraphfor defining workflow nodes and edges - Implements
Pregelfor distributed graph processing
- Uses
- LangChain: Provides the foundation for working with LLMs
- Uses
BaseChatModelfor standardized LLM interactions - Implements message handling with
SystemMessageandHumanMessage
- Uses
- Pydantic: Handles data validation and settings management
- Implements
BaseModelfor type-safe data structures - Provides schema validation for API inputs and outputs
- Implements
- AsyncResult: Custom implementation for handling asynchronous tasks
- Manages polling and result retrieval from external APIs
- Implements timeout and error handling mechanisms
- NVIDIA Cosmos API: Generates 3D scenes from text descriptions
- Three.js: Creates interactive 3D visualizations in the browser
- Prompt Upsampling: Enhances simple text descriptions into detailed scene descriptions
- Asynchronous Processing: Handles video generation requests asynchronously
- Error Handling: Gracefully handles failures with fallback options
NVIDIA_API_KEY: Required for authenticating requests to the NVIDIA Cosmos API
The documentation and code for this project were generated and enhanced using Windsurf, an agentic IDE powered by Codeium's AI Flow paradigm.
To use this backend service, ensure you have the required API keys set in your environment and follow the API documentation for making requests to the service.