ChrisChen667788/wind-comic

Multi-agent AI pipeline that turns one line of text into a finished short-form drama: script, cinematic storyboards, character-consistent video. Provider-agnostic (OpenAI/Claude, MJ, Minimax, Veo/Sora, fal, ComfyUI). MIT.

What it solves

Most AI video tools only generate short, disconnected clips from a single prompt. Wind Comic solves the problem of end-to-end production by turning a single line of text into a complete, cohesive short-form drama, including scripts, character bibles, storyboards, voiceovers, and a final MP4 file.

How it works

The system uses an honest multi-agent pipeline where specialized agents (Writer, Director, Character Designer, etc.) hand off tasks with strict consistency contracts. It follows a structured workflow: Text $\rightarrow$ JSON $\rightarrow$ PNG $\rightarrow$ IMG $\rightarrow$ MP4. The process includes a "Vision-Audit" retry loop that automatically regenerates shots scoring below a certain threshold and a "multi-engine race" where multiple video engines compete to provide the best clip.

Who it’s for

Short-drama creators looking for trope-aware writing and cliffhanger detection.
Content marketing teams needing consistent characters across polished 30-second ads.
Indie filmmakers requiring a multi-track timeline and cinematography control.
Comic adaptation studios needing to convert scripts to storyboards with consistent art styles.
Open-source builders who want to plug in their own LLMs (OpenAI, Anthropic, DeepSeek, Ollama, etc.).

Highlights

Multi-agent orchestration involving specialized roles like Director, Writer, and Editor.
Character consistency via a "Character Studio" that locks facial geometry and identity DNA.
End-to-end lip-sync with viseme keyframe tracks and mouth-to-audio alignment scoring.
Director's control room featuring a multi-track timeline and cinematography controls.
Bring-your-own LLM support via simple environment variables.
Automated quality loops for re-rendering degraded shots and auditing film health.