wind-comic  by ChrisChen667788

AI multi-agent pipeline crafts complete short-form dramas from a single text prompt

Created 2 months ago
320 stars

Top 84.4% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

Wind Comic is a multi-agent AI pipeline that transforms a single text prompt into a complete short-form drama, addressing the gap between short AI video clips and finished productions for creators and marketers. It offers a provider-agnostic solution for generating scripts, storyboards, and character-consistent video.

How It Works

Wind Comic employs a multi-agent architecture where specialized AI roles (Writer, Director, etc.) collaborate. Each agent adheres to strict consistency contracts, handing off artifacts stage-by-stage. This modular approach allows for flexibility, independent re-running of stages, and seamless integration of diverse LLMs and media engines, avoiding the limitations of monolithic AI models.

Quick Start & Requirements

Install via npm install after cloning. Configure essential LLM API keys and base URLs in .env.local. Requires Node.js and a JSON-responding LLM (≥24B params). Optional API keys for various media engines are supported. See docs/llm-providers.md for details.

Highlighted Details

  • Multi-agent pipeline orchestrates 8 specialized agents for end-to-end short-form drama production (script to MP4).
  • "Style Bible Frame" ensures visual coherence across all shots by using a canonical key art frame as a reference.
  • Trope-aware generation defaults to 9:16 vertical canvas with 12 short-drama templates and hook patterns.
  • Real CJK subtitle burning integrates subtitles correctly by stripping text from video prompts and post-baking with system CJK fonts.
  • Character consistency is achieved via cref, sref, 8-dimension DNA extraction, and Vision Retry.
  • Real-time collaboration features a multi-track timeline with Yjs CRDT for live multiplayer editing.
  • Advanced lip-sync pipeline includes per-character voice routing, viseme alignment, and pluggable engines.
  • Pacing audit analyzes shots for conflict, reversals, and cliffhangers, offering actionable suggestions.
  • Bring Your Own LLM (BYO LLM) allows seamless swapping of LLM providers via environment variables.
  • Robustness is ensured by 1894 unit tests and strict TypeScript typing.

Maintenance & Community

The project emphasizes contributions via PRs and robust testing. Roadmap and version history are available in ROADMAP.md and VERSIONS.md respectively. No explicit community channels (e.g., Discord/Slack) are listed.

Licensing & Compatibility

Licensed under MIT, permitting commercial use, forking, and building startups. Contributions via PR are encouraged.

Limitations & Caveats

Requires specific LLM configurations (JSON response, ≥24B params). Higher video concurrency may weaken keyframe-chain continuity. Integration of optional media engines relies on API key availability and graceful fallback mechanisms.

Health Check
Last Commit

22 hours ago

Responsiveness

Inactive

Pull Requests (30d)
1
Issues (30d)
1
Star History
200 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.