An MCP-based AI video editing agent that enables full-pipeline video creation—from material search and script generation to final rendering—through natural language interaction.
FireRed-OpenStoryline is a conversational AI video editing agent built on an MCP Server + LangChain Agent architecture, transforming traditional manual editing into intent-driven video creation.
The project covers the full video production pipeline: intelligent material search (Pexels), shot segmentation (TransNetV2), VLM-based visual understanding, few-shot stylized script generation, multi-provider TTS voiceover, emotion-driven BGM recommendation with beat synchronization, AI transition generation (Tongyi Wanxiang/Hailuo), and final rendering via MoviePy + FFmpeg. It includes 18 processing nodes exposed through an MCP Server using the streamable-http protocol.
Interaction is available via both CLI and Web UI, with conversational commands for shot editing, replacement, reordering, and visual detail fine-tuning. The Style Skills mechanism allows saving complete editing workflows as reusable skills for batch stylized production through material replacement. It also provides Skills integration for AI agents like OpenClaw, Claude Code, and Codex, enabling end-to-end automation from installation to generation.
Supports local installation (Python >= 3.11) and Docker deployment, cross-platform (Linux/macOS/Windows), with external LLM/VLM and related API keys required.
Note: The README contains date annotations from 2026 (open-source date and new feature dates), which may be future dates or internal planning timelines in the project documentation. Actual dates should be verified via Git history. The specific URLs for the HuggingFace Demo and project Homepage were not captured during research.