ChrisChen667788/wind-comic

Wind Comic – AI‑Powered Short‑Form Drama Studio

One line of text → a complete short‑film (script, storyboard, voice‑over, lip‑sync, music, final .mp4). Wind Comic is a multi‑agent, web‑based production pipeline that stitches together a collection of specialised AI models (LLM, image diffusion, video engines, TTS, lip‑sync) under a single collaborative interface.

What it does

• Input: a single sentence (or a short prompt).
• Output: a finished short‑form drama – full script, character bible, multi‑shot storyboards, per‑character voice‑over, synchronized mouth animation, background music, subtitles and an MP4 video.
• Workflow: a chain of autonomous agents – Writer, Director, Producer, Character Designer, Storyboard Artist, Video Engine, TTS, Lip‑Sync, Editor – each receives a strict contract and passes artefacts to the next stage. The system stores every intermediate artefact (JSON, PNG, MP4) in a dual‑driver database (SQLite ↔ PostgreSQL) so any stage can be re‑run without re‑generating the whole film.
• Real‑time collaboration: a CRDT‑based timeline (Yjs) lets multiple users edit the project simultaneously, similar to Figma for video.
• Bring‑Your‑Own LLM: you can swap the default LLM (DeepSeek → MiniMax) by setting three environment variables – no code changes required.
• Plug‑in media engines: 12+ image/video generators (Midjourney, Seedream, Kling, Veo, HappyHorse, etc.) are routed through a configurable priority list; the first engine that returns a satisfactory clip wins the “engine race”.
• Cost guard & budgeting: per‑stage cost attribution (LLM, image, video, TTS, lip‑sync) with budget thresholds and a four‑dimensional publish gate (visual‑script consistency, lip‑sync quality, subtitles, budget).
• Template market: finished projects can be saved as reusable templates (style, pacing, voices) and remixed with a single click.

Key technical highlights

• Frontend: Next.js 16 (React) with a WebGL‑based director console for camera placement, 3‑D preview, and tilt/aim controls.
• Backend orchestration: Node 20 server orchestrates the eight agents, handles CRDT sync, and talks to the LLM gateway and media‑engine router.
• Data layer: SQLite for local dev, PostgreSQL for production; all artefacts are persisted and versioned.
• Quality loops: Vision‑audit retry (auto‑regenerate shots scoring < 70), multi‑engine race, ffprobe‑driven health reports, and a viseme‑vs‑audio alignment score that drives automatic lip‑sync drift correction.
• Testing: 6608 automated tests pass in CI, covering the full pipeline.
• Internationalisation: UI and generated scripts support both English and Chinese; subtitles are auto‑generated per‑project language.

Who might use it

• Content creators who want to produce short narrative videos without hiring a full production crew.
• Game or visual‑novel studios looking to prototype cut‑scenes from a novel or script quickly.
• Educators & marketers needing rapid, stylised video explanations or ads.
• Teams that need collaborative story‑boarding and version control for video assets.

Maturity & Availability

• The project is at version v13.4 (released Oct 2026) with a full changelog, CI pipeline, and a public demo video.
• Licensed under MIT, with a Chinese mirror on AtomGit for mainland users.
• Documentation includes a pitch deck, LLM‑provider guide, and a roadmap.

---

Bottom line: Wind Comic is a production‑grade, multi‑agent AI studio that turns a single line of text into a complete short‑film, offering real‑time collaboration, cost visibility, and extensibility via plug‑in LLMs and media engines.