The 9-Agent Pipeline
Each agent specializes in one stage of the video production workflow, orchestrated through a LangGraph state graph.
Research Agent
Searches the web to inspect known stories, references, products, and brands to write high-fidelity background notes.
Learn more →Script Agent
Consumes research notes to write cohesive, structured outlines and spoken voiceover scripts without copy-pasting source texts.
Learn more →Art Style Agent
Selects from 34 styles like Neon Noir, Watercolor Anime, or Documentary Realism to match the narrative tone.
Learn more →Storyboard Agent
Translates the script into sequential scenes, defining exact shot compositions, camera angles, and character actions.
Learn more →Character Agent
Generates character model descriptions and consistency parameters, matching them to appropriate vocal profiles.
Learn more →Location Agent
Designs environment key art and background plates, ensuring consistency for all actions in a scene.
Learn more →Voice Agent
Generates expressive, multi-character voiceovers with fine-tuned gender, accent, and emotional matching.
Learn more →BGM Agent
Analyzes the script's emotional curve to compose or select matching background soundtracks and audio transitions.
Learn more →Media Agent
Synthesizes images, voices, and audio assets, feeding them to the render worker to compile the final video.
Learn more →Pipeline Architecture
OpenDirector's 9 agents are orchestrated through a LangGraph state graph. Each agent is a node that receives output from upstream agents as input, processes it, and passes results downstream. This architecture supports real-time collaboration, checkpoint recovery, and manual intervention at any stage. The pipeline starts from a single sentence prompt, flowing through research, scripting, art style selection, storyboard planning, character and location design, voice generation, music composition, and finally the Media Agent synthesizes all assets into a complete rendered video.