// the find
FireRedTeam/FireRed-OpenStoryline
FireRed-OpenStoryline is an AI video editing agent that transforms manual editing into intention-driven directing through natural language interaction, LLM-powered planning, and precise tool orchestration. It facilitates transparent, human-in-the-loop creation with reusable Style Skills for consistent, professional storytelling.
An LLM-orchestrated video editing agent: you describe what you want in natural language, and it drives a pipeline of nodes (ASR, shot splitting, timeline planning, rendering) via MCP/LangChain to produce a cut video, with 'Skills' letting you save a workflow and replay it on new footage. Aimed at people doing repetitive short-form content (vlogs, product reviews, unboxings) who want batch output in a consistent style rather than manual timeline editing.
The node-based architecture (search_media, understand_clips, plan_timeline, render_video, etc.) is a real pipeline decomposition, not just a chatbot wrapper around ffmpeg. The Skill system is a legitimate product idea: a saved SKILL.md plus script encodes an entire editing workflow so you can swap media and get the same style back, which is more durable than one-off prompting. Prompt engineering is organized per task and per language (en/zh) across a dozen task folders, suggesting actual iteration rather than a single mega-prompt. Ships multiple entry points (CLI, FastAPI web server, Docker image) so it's not locked into one integration path.
The default asset library is explicitly labeled 'Restricted Mode' in the README - decent fonts/music/VFX require a proprietary Xiaohongshu asset library, so out-of-the-box output quality is intentionally capped. The flagship AI transition feature depends on third-party AIGC video generation services that the README itself warns are expensive and produce 'somewhat unpredictable' results. There are no tests anywhere in the tree despite orchestrating rendering and LLM tool-calling, which is exactly the kind of pipeline that breaks silently. Windows setup means manually downloading and unzipping two asset bundles from a Tencent COS URL rather than a proper release artifact, which is a rough first-run experience outside Linux/macOS.