An AI film studio that blocks actors before burning your credits

What it does Wind Comic is a TypeScript/Next.js studio that turns a single sentence into a complete short-form drama — script, reusable character bible, multi-shot storyboards, per-character voiceover, background music, and a final cut. It acts as an orchestration layer across LLMs and generative media providers, routing work through specialist agents (Writer, Director, Storyboard Artist, Editor, and others) that hand off artifacts with strict consistency contracts. The whole thing is wrapped in a real-time collaborative timeline that lets a team edit like Figma for film.

The interesting bit Instead of treating video generation as a lottery, the pipeline includes a Director’s console where you block actors and camera geometry before spending money; a deterministic composition audit flags who is out of frame before a single provider call is made. If a shot still degrades, the system can retake just the broken two seconds by byte-copying the rest, and it races multiple video engines — Kling, Veo, MiniMax, Seedance — to pick the first acceptable clip rather than praying over one.

Key highlights

• Eight specialist agents handle distinct production roles, from casting to lip-sync, with reusable character assets and a shared asset store.
• Sketch-lock storyboarding forces diffusion models to respect a pre-approved B/W composition, so camera language survives the generation lottery.
• A built-in quality loop runs ffprobe health reports on every shot, auto-rerenders degraded frames, and exports a labeled animatic rather than pretending a failed still image is video.
• Bring-your-own-LLM and bring-your-own-provider setup: swap OpenAI, Claude, DeepSeek, Midjourney, ComfyUI, or Kling via environment variables without code changes.
• Per-project cost attribution and budget guards try to keep the multi-engine race from becoming a multi-engine money pit.

Caveats

• Photorealistic lip-sync is not included out of the box; you need to wire up external tools like wav2lip, SadTalker, MuseTalk, or Kling yourself.
• The project is essentially a heavy orchestration layer across many paid third-party APIs, so running it means managing keys, rate limits, and bills across a dozen services.
• The toolset is heavily opinionated toward short-form drama workflows (including CJK subtitle support and season-splitting), which may feel prescriptive if you just want a generic video generator.

Verdict Teams already paying for multiple generative media APIs and tired of manual prompt-hunting for character consistency should look here. If you want a single self-hosted model that works offline, this is the wrong stack — it’s glue, not a monolith.