A film director agent breaks a story into visual scenes. Each scene gets a script excerpt and a Stable Diffusion prompt. BatchSplitter separates the scenes for downstream image generation.
Story concept → LLM writes scenes → ToolParser saves characters → BatchSplitter outputs individual scenes.
Splits the LLM output on ## Scene headers. Returns the first scene individually plus a JSON array of all scenes — perfect for feeding into image generation loops.
The director stores character descriptions and scene scripts in memory. Next time you run the workflow, new scenes reference existing character designs for visual consistency.
Each scene comes with a script excerpt AND a Stable Diffusion prompt. Wire the scene outputs directly into your image generation pipeline.