通用 AI 视频制作多 agent 模板 / Universal AI Video Production Multi-Agent Template
Agentic Video Studio 是一个覆盖全视频类型的 agentic AI 视频制作中枢。从一句创意出发，把脚本、storyboard、AI 生成镜头、配音、配乐、动画、剪辑、字幕、缩略图、多平台分发，全部交给 50+ 协同 agent 自动完成。无论是长视频、短视频、课程录播、产品演示、播客视频、动画解说、数据可视化视频、白板动画、MV、预告片、采访、短剧——都跑同一个工作室，只是路由不同。
- 50+ 专业 agent：每个负责一段制作流程，全部用 markdown 描述提示词，所有模型可换可调可读。
- 80+ skill：从"生成 voiceover"这样的原子操作，到"完整长视频生产"这样的端到端工作流，分层组合。
- 15+ MCP server：ElevenLabs、HeyGen、Runway、Pika、Kling、Luma、Hailuo、Sora、Suno、Udio、AssemblyAI、YouTube、Bilibili、TikTok、Manim、Remotion、Hyperframes 全部集成。每个都带 stub fallback——没 API key 也能跑通整条流水线，方便先把流程打通再付费。
- YOLO 模式 + 花费台账：一行
/yolo启动全自动模式，每一笔 API 消耗自动写入spend-ledger.jsonl，每月可审计。 - Headshot Studio 集成：talking-head 段落直接调用姊妹项目 Agentic-Headshot-Video-Studio 的成熟 agent 集，无重复造轮子。
- 完整 onboarding：一句
/onboard从视频类型、目标平台、风格、自动化等级，到 API key 收集、MCP 安装、self-test，全部走完。
| 场景 | 推荐工作流 | 主力 agent |
|---|---|---|
| YouTube 长视频教程 | full-long-video |
script-writer-long / b-roll-curator / multicam-editor / seo-optimizer / youtube-uploader |
| 抖音/视频号/Reels 短视频 | full-short-video |
script-writer-short / talking-head-specialist / vertical-formatter / tiktok-uploader |
| 课程录播 | educational-course-module |
lesson-planner / lecture-segmenter / subtitle-generator / chapter-marker |
| 产品演示 | product-demo |
screen-recorder-orchestrator / callout-animator / lower-third-designer |
| 播客视频版 | podcast-video |
multicam-editor / audio-engineer / waveform-visualizer / chapter-marker |
| 动画解说视频 | animated-explainer |
motion-graphics-artist / 2d-animator / kinetic-typographer |
| 数学/编程可视化 | code-visualization |
mathematical-visualizer-manim / data-viz-animator |
| 数据故事视频 | data-storytelling |
data-viz-animator / chart-narrator / kinetic-typographer |
| MV / 音乐视频 | music-video |
music-composer / beat-syncer / motion-graphics-artist |
| 预告片 / 宣传片 | trailer-promo |
hook-writer / sfx-designer / pacing-reviewer |
| Vlog | travel-vlog / food-vlog |
rough-cutter / color-grader / transition-specialist |
| 短剧 / 喜剧 skit | skit-comedy |
script-writer-skit / character-designer / dialogue-balancer |
git clone https://github.com/huodebing-alt/Agentic-Video-Studio.git
cd Agentic-Video-Studio
claude # 或者你的任何 Claude Code / Antigravity / Cursor / Cline 客户端
# 在对话里输入：
/onboard/onboard 会引导你：
- 选视频类型（长 / 短 / 课程 / 演示 / 动画 / 等 18 种）
- 选目标平台（YouTube / Bilibili / TikTok / 视频号 / 小红书 / IG / LinkedIn）
- 选风格（严肃 / 教学 / 综艺 / 商务 / 极简）
- 选自动化等级（人工逐步审核 / YOLO 全自动）
- 收集需要的 API key（带链接和价格说明）
- 安装并 self-test 对应 MCP server
| 项目 | 定位 | 何时用 |
|---|---|---|
| Agentic-Video-Studio（本项目） | 通用视频制作 hub | 任何视频类型——本项目能 cover |
| Agentic-Headshot-Video-Studio | Talking-head 专项 | 只做 headshot/讲解人视频，需要 lighting/wardrobe/teleprompter 等精细 agent |
两个项目设计上互不替代：Headshot Studio 提供 talking-head 专项的精细 agent 集，本项目把它当作其中一条路径调用，不重写。详见 docs/RELATED_PROJECTS.md。
所有 agent 都是 markdown 提示词文件，没有写死任何模型。你可以：
- Claude Opus / Sonnet / Haiku
- GPT-4 / GPT-5 / o3
- Gemini 2.5 Pro / Flash
- DeepSeek / Qwen / GLM / Kimi / Doubao
- 任意本地模型（Llama / Qwen / DeepSeek-Coder）
只要客户端支持读 .claude/agents/*.md，就能跑。
如果这个项目帮你省下了至少一周的工程时间，欢迎打赏作者一杯咖啡：ko-fi.com/huodebing / GitHub Sponsors / 爱发电。
我做开源不是为了挣钱，是为了让全世界的小团队都能做出和大厂一样的视频。但如果你能给我买杯咖啡，我会更有动力把下一个 hub 做出来——agentic 直播间、agentic 综艺剪辑、agentic AI 主播。让我们看看一个人 + 一群 agent 能把内容生产推到什么程度。
Agentic Video Studio is an agentic AI video production hub covering every video type. From a single idea to a finished video on every major platform: scripting, storyboard, AI-generated shots, voiceover, music, animation, editing, subtitles, thumbnails, multi-platform distribution—all handled by 50+ coordinated agents. Long-form, short-form, course recordings, product demos, podcast videos, animated explainers, data-viz videos, whiteboard animation, music videos, trailers, interviews, skits—all run the same studio with different routing.
- 50+ specialized agents: each owns a slice of the production pipeline. All written as markdown prompts—model-swappable, model-agnostic, human-readable.
- 80+ skills: from atomic operations like "generate voiceover" to end-to-end workflows like "produce a full long-form video". Layered, composable.
- 15+ MCP servers: ElevenLabs, HeyGen, Runway, Pika, Kling, Luma, Hailuo, Sora, Suno, Udio, AssemblyAI, YouTube, Bilibili, TikTok, Manim, Remotion, Hyperframes. Each ships with a stub fallback—your pipeline runs end-to-end with zero API keys, so you can validate logic before paying.
- YOLO mode + spend ledger: one
/yoloflips the studio into fully autonomous mode; every API call is appended tospend-ledger.jsonlfor monthly audit. - Headshot Studio integration: talking-head segments delegate to the sister project Agentic-Headshot-Video-Studio—no reinvention.
- Full onboarding: a single
/onboardwalks you through video type → platform → style → automation level → API key collection → MCP install → self-test.
git clone https://github.com/huodebing-alt/Agentic-Video-Studio.git
cd Agentic-Video-Studio
claude # or any Claude Code / Antigravity / Cursor / Cline client
# In the chat:
/onboard| Project | Scope | When to use |
|---|---|---|
| Agentic-Video-Studio (this) | General video production hub | Any video type—this project covers it |
| Agentic-Headshot-Video-Studio | Talking-head specialty | When you only do headshot/explainer-person videos and need fine-grained agents for lighting/wardrobe/teleprompter |
They are designed to complement, not replace each other.
If this project saved you a week of engineering, buy me a coffee: ko-fi.com/huodebing.
I don't open-source for the money—I open-source so a one-person team can ship videos that look like a studio made them. But if you can spare a coffee, I'll have more energy to ship the next hub: an agentic livestream room, agentic variety-show edit, agentic AI host. Let's find out what one human + a swarm of agents can do to content production.
License: MIT · Author: @huodebing-alt · Issues / PRs welcome