AutoVideo-Agent

Turn a Markdown script into a reproducible video pipeline — storyboard, scene assets, timeline, QA, and MP4.

Built for Codex, Claude Code, Gemini CLI and other coding-agent workflows. v0.1 is local-first and deterministic: it creates inspectable placeholder scene assets and an FFmpeg video without an API key or cloud account.

Markdown Script -> Storyboard -> Scene Manifest -> Media -> Timeline -> FFmpeg -> QA -> MP4

The default v0.1-compatible command does not claim AI video generation. v0.2 provider mode adds ComfyUI API media, while MiniMax and other hosted providers remain planned.

Demo

The demo uses examples/demo-script.md: three scenes and seven seconds.

INPUT                         PIPELINE                         OUTPUT
examples/demo-script.md  ->  autovideo run              ->  video.mp4
                              parse + manifest + assets      manifest.json
                              silent WAV + FFmpeg            report.json

Run it:

autovideo run examples/demo-script.md

The build is written to build/demo-script/. Open video.mp4 when FFmpeg is available. Always inspect manifest.json and report.json; without FFmpeg the command reports status: degraded and keeps the inspectable assets.

Quick Start

Install the package from PyPI:

python -m pip install autovideo-agent

To run the repository demo, clone the repository for its example script:

git clone https://github.com/wangxin6x/AutoVideo-Agent.git
cd AutoVideo-Agent
autovideo run examples/demo-script.md

The wheel contains the autovideo CLI and runtime package. examples/ is a repository fixture, so use your own Markdown script after installing from PyPI or clone the repository to run this demo.

FFmpeg is optional. With it, the output is an H.264 MP4 with a silent AAC track. Without it, scene cards, manifest, WAV timeline, and QA report are still produced.

Features

Status	Capability	Evidence
✅ Available now	Markdown storyboard parser	src/autovideo/parser.py
✅ Available now	Scene manifest	manifest.json
✅ Available now	Deterministic offline assets	PPM scene cards
✅ Available now	Silent WAV timeline	audio-silence.wav
✅ Available now	FFmpeg MP4 rendering	src/autovideo/render.py
✅ Available now	Graceful degradation	report.json status
✅ Available now	CLI	autovideo run <script.md>
✅ Available now	QA report	report.json
✅ Available now	Codex Skill / AGENTS integration	AGENTS.md and skills/auto-video/SKILL.md
🧪 Experimental	ComfyUI API media provider	Implemented; API workflow submit, poll, retry, resume, and download; awaiting live validation
✅ Available now	Mock and command TTS providers	Silent fallback or any local TTS CLI
✅ Available now	Scene-level SRT subtitles	Timed from actual TTS audio duration
🚧 Planned	MiniMax	#1
🚧 Planned	Hosted TTS integrations	OpenAI, Volcengine, and ElevenLabs
🚧 Planned	Word-level subtitle alignment	#4
🚧 Planned	Real media adapters	#5

Architecture