waoowaoo

Orchestrates a sequential workflow that transforms novel text through six distinct stages: configuration, script generation, asset creation, storyboard composition, video synthesis, and voice-over production. Uses a graph runtime system with event-driven task submission to coordinate LLM calls, image generation, video synthesis, and voice synthesis across multiple AI providers, with React Query managing client-side state synchronization and background task polling.

Accepts novel text and generates screenplays/scripts using configurable LLM providers (OpenAI, Anthropic, etc.) through an abstraction layer that handles model selection, prompt engineering, and output parsing. The system maintains provider configuration state and billing tracking per model, allowing users to switch between providers and models without code changes. Integrates with the task infrastructure to submit LLM tasks asynchronously and track completion via event system.

Manages the lifecycle of generated artifacts (images, videos, audio files) with versioning, reference tracking, and cleanup policies. The system tracks which artifacts are used in which stages (e.g., character image used in storyboard frame), prevents deletion of in-use artifacts, and maintains artifact metadata (generation parameters, provider, timestamp). Implements a media reference system that maps artifacts to their usage locations in the project.

Implements workspace-level isolation that separates projects, assets, and credentials between different users or teams. The system enforces access control at the workspace level, with role-based permissions (admin, editor, viewer) for project access. Each workspace maintains its own Asset Hub, project list, and provider configurations, with no cross-workspace data sharing except through explicit export/import.

Generates character images and location backgrounds using image generation APIs (Midjourney, DALL-E, Stable Diffusion) with style reference forwarding to ensure visual consistency across all generated assets. The system maintains a character management subsystem that stores character descriptions, appearance references, and style parameters, then injects these into image generation prompts. Uses a candidate selector UI that presents multiple generation options for human approval before committing assets to the project.

Composes storyboards by sequencing generated character and location assets into frames that correspond to screenplay scenes. The system maps screenplay scenes to storyboard frames, selects appropriate character and location assets for each frame, and presents a visual timeline for human review and editing. Uses a frame-level candidate selector that allows swapping assets, reordering scenes, or adjusting frame timing before committing to video synthesis.

Synthesizes animated videos from storyboard frames and voice-over audio using video generation APIs (Runway, Synthesia, or equivalent) with integrated lip-sync to match character mouth movements to dialogue. The system submits video synthesis tasks asynchronously, tracks generation progress, and returns final video files with synchronized audio and animation. Handles frame-to-frame transitions and character positioning based on storyboard layout.

Synthesizes voice-over audio from screenplay dialogue using text-to-speech APIs (ElevenLabs, Google Cloud TTS, Azure Speech, etc.) with character-to-voice assignment and voice cloning support. The system maintains a voice management subsystem that stores voice profiles (provider, model, language, tone), maps characters to voices, and generates audio for each dialogue line. Supports voice cloning from reference audio samples to create custom character voices.

Maintains a global asset library (Asset Hub) that stores reusable character definitions, location backgrounds, and voice profiles across all projects. The system allows users to create global assets once and reference them in multiple projects, with project-level asset overrides for customization. Uses a hierarchical asset management system that separates global assets (shared across workspace) from project assets (specific to one project), with asset versioning and usage tracking.

Implements an asynchronous task queue system that submits image generation, video synthesis, LLM, and voice synthesis tasks to background workers with provider-specific handlers. The system maintains task state (pending, running, completed, failed), tracks task progress, and provides retry logic with exponential backoff. Uses event-driven architecture where task completion triggers downstream stage transitions and artifact management updates.

Uses React Query to manage client-side state for projects, assets, tasks, and workflow progress with automatic background polling for task status updates. The system maintains query caches for project data, asset lists, and task status, with mutations for creating/updating projects and submitting tasks. Implements polling intervals that adapt based on task state (faster polling for in-progress tasks, slower for completed tasks) to balance responsiveness and server load.

Provides a configuration system that allows users to select and configure multiple AI providers (LLM, image generation, video synthesis, TTS) at the project level, with secure credential storage and per-provider model selection. The system validates API credentials on configuration and tracks provider usage and costs per project. Supports switching providers mid-project without losing project state, with automatic provider failover if configured.