UniVA is an open-source system for planning, composing, and producing videos via natural language instructions, acting as an intelligent video director. It targets users seeking an agentic, proactive workflow for video creation, enabling iterative co-creation and complex narrative generation. The primary benefit is an end-to-end workspace that simplifies video production through an AI-driven director.

How It Works

UniVA employs an agentic creation approach featuring multi-round co-creation, deep memory for context consistency, implicit intent reading, and proactive suggestions. Its architecture comprises a Python backend with a Plan Agent for high-level task decomposition and an Act Agent for execution, supported by Modular Context Protocol (MCP) tools for video processing and generation. A FastAPI server provides a RESTful API, while an optional Next.js frontend offers a web interface and video editor. The system supports a universal video fabric, accepting various inputs (text, image, entity, video) for controllable video output, with capabilities for super HD, cinematic quality, complex narratives, and fine-grained editing.

Quick Start & Requirements

• Primary Install: Clone the repository, then install Python dependencies using pip install -r requirements.txt or pip install -e ..
• Prerequisites: Python 3.10+, Node.js 18.0+ (for frontend), Bun 1.2.18+ (for frontend), CUDA (recommended for GPU acceleration).
• Configuration: Requires editing configuration files to set API keys (e.g., OpenAI for planning/acting agents) and MCP server paths.
• Usage: Backend can be run interactively via CLI (python univa/univa_agent.py) or as a server (python univa_server.py) for API/web access. Frontend development server runs with bun run dev.
• Links: Project repository: https://github.com/univa-agent/univa.

Highlighted Details

• Agentic Creation: Features multi-round co-creation, deep memory, implicit intent reading, and a proactive agent that suggests better shots and stories.
• Omnipotent Video Production Engine: Supports a universal video fabric (Text/Image/Entity/Video → controllable video), any-conditioned pipeline, complex narratives, and ultra-long/fine-grained editing.
• Extensibility: MCP-native modular design facilitates easy extension with new models and tools.

Maintenance & Community

Contributions are welcomed across various areas including bug fixes, new features, documentation, and UI/UX improvements. The project acknowledges the OpenCut project and the broader open-source community. Specific community channels (e.g., Discord, Slack) or a public roadmap are not detailed in the README.

Licensing & Compatibility

The README does not explicitly state the project's license. This omission requires further investigation for users considering commercial use or integration into closed-source projects.

Limitations & Caveats

No explicit limitations or caveats are mentioned in the provided README. The project requires configuration of external API keys (e.g., OpenAI) for core agent functionality. The optional frontend has additional Node.js and Bun dependencies.