An open-source AI video studio with a 9-agent director pipeline — from a one-line idea to a fully rendered video with voiceover, BGM, and storyboard.

English | 中文

🌐 Official Website: https://od.seme.cc

OpenDirector is a Docker-first, self-hosted AI video production studio. Describe your idea in one sentence, and a team of 9 specialized AI agents collaborate to produce a complete video — with optional web research, a storyboard, character designs, voiceover, background music, and rendered output.

Just docker compose up and start creating.

AI Director Chat	Batch Production
Creation Editor	Storyboard Preview

Your Idea
   |
   v
[Research Agent] --> [Script Agent] --> [Art Style Agent] --> [Storyboard Agent]
                                                            |
                                                            v
[Character Agent] --> [Location Agent] --> [Voice Agent] --> [BGM Agent]
                                                            |
                                                            v
                                                     [Media Agent]
                                                            |
                                                            v
                                                  [Render Worker] --> Final Video

9 specialized agents work in a pipeline:

1. Research Agent — uses OpenAI web_search_preview when needed to check known stories, factual references, brands, products, and source notes
2. Script Agent — generates the story outline and narrative structure, using research notes when available
3. Art Style Agent — selects from 34 built-in styles (e.g. Futuristic Neon Noir, Dreamscape Watercolor Anime, Documentary Realism)
4. Storyboard Agent — breaks the story into scenes with shot descriptions and dialogue
5. Character Agent — designs characters with visual prompts and assigns voice profiles
6. Location Agent — creates environment concepts for each scene
7. Voice Agent — assigns TTS voices matched to character personality and gender
8. BGM Agent — generates background music based on story atmosphere
9. Media Agent — orchestrates image/voice/music generation into final assets

Each agent is a LangGraph node that streams its output in real-time — you can watch the plan build step by step. The shared graph state now includes a research field with notes, cautions, and sources. The Script Agent consumes those notes without copying source text.

• Input one sentence, AI director auto-generates complete plan: brief, story, storyboard, voiceover, images, BGM
• Optional web research for known stories, factual references, brands, products, and public information
• 34 built-in art styles across 9 categories: Cinematic, Commercial, Futuristic, Retro, Anime, 3D, Illustration, Realistic, Experimental
• AI-generated story scripts, editable manually
• AI voiceover with multiple voice options, real-time preview
• AI background music, auto-generated based on story atmosphere
• Storyboard preview with image + voiceover + BGM synced playback
• Support 16:9 / 9:16 / 1:1 aspect ratios
• Export at 480p / 720p / 1080p

• Input topics, AI auto-generates multiple scripts, batch produce short videos
• Configurable clip duration (2-10 seconds), control material switching rhythm
• Support Chinese and English video scripts
• Multiple TTS voices with built-in Edge TTS (free), real-time preview
• Subtitle generation with customizable font, size, color, position, stroke
• Background music — random or specified local files, adjustable volume
• Video materials are HD and royalty-free (Pexels / Pixabay), local files also supported
• Generate multiple output variations at once, pick the best one

• Multiple AI model providers — OpenAI, Google Gemini, DeepSeek, Qwen, MiniMax, Ollama, and more
• Pluggable media providers — AiHubMix, WaveSpeed, switch via environment variable
• Docker one-click deploy — docker compose up and you're ready
• Fully self-hosted — data stays on your server
• Chinese and English UI

• Docker & Docker Compose
• (Optional) Node.js 20+ for local development

git clone https://github.com/seme-org/open-director.git
cd open-director
cp .env.example .env
# Edit .env with your API keys
docker compose up --build

Then open http://localhost:3000.

OpenDirector uses WaveSpeed for image generation. Character and location plates use a text-to-image model, while storyboard frames with character references must use an image-to-image/edit model so reference images are honored.

WAVESPEED_API_KEY="your-wavespeed-key"
WAVESPEED_IMAGE_MODEL="nano-banana"
WAVESPEED_IMAGE_TO_IMAGE_MODEL="nano-banana-2-edit"
EDGE_TTS_VOICE="zh-CN-XiaoxiaoNeural"

Do not set WAVESPEED_IMAGE_TO_IMAGE_MODEL to nano-banana: that alias routes to a text-to-image endpoint and can ignore character reference images.

Speech uses local Edge TTS. Background music uses local tracks from assets/bgm/default/.

The LLM is used for recipe generation, script writing, and the AI director. It uses OpenAI-compatible API format.

OPENAI_API_KEY="your-key"
OPENAI_BASE_URL="https://api.openai.com/v1"
OPENAI_MODEL="gpt-4o-mini"

Use OpenAI directly or an OpenAI-compatible endpoint for the LLM.

pnpm install
pnpm db:generate
pnpm dev

This starts the Next.js dev server on http://localhost:3000.

┌─────────────────────────────────────────────────┐
│                   Next.js App                   │
│  (App Router, React 19, TypeScript, Tailwind)   │
└────────┬──────────┬──────────┬──────────────────┘
         │          │          │
    ┌────▼───┐ ┌────▼───┐ ┌───▼────┐
    │ MySQL  │ │ Redis  │ │ MinIO  │
    │   8.4  │ │   7    │ │  S3    │
    └────────┘ └────┬───┘ └────────┘
                    │
              ┌─────▼──────┐
              │   Worker   │
              │ (FFCreator) │
              └────────────┘

open-director/
├── apps/
│   ├── web/          # Next.js frontend + API routes + 9 AI agents
│   └── render/       # BullMQ render worker (FFCreator)
├── assets/
│   └── fonts/        # Subtitle rendering fonts
├── prisma/
│   └── schema.prisma # Database schema (voices, art_styles, bgms, etc.)
├── docker-compose.yml
└── package.json

apps/web/src/server/agent/
├── media-provider.ts          # Types + factory + orchestrator
├── schemas/
│   └── research.ts            # Research notes, cautions, and source schema
├── voices.ts                  # TTS voice catalog (loaded from database)
├── art-styles.ts              # Art style catalog (loaded from database)
├── providers/
│   ├── wavespeed.ts           # WaveSpeed implementation
│   ├── aihubmix.ts            # AiHubMix implementation
│   ├── local-bgm.ts           # Local BGM (random track from database)
│   └── wavespeed.test.ts      # Provider tests
└── graph/nodes/recipe/        # 9 LangGraph agent nodes

docker compose up -d --build

This starts all services: MySQL, Redis, MinIO, web app, and render worker.

pnpm install
pnpm db:generate
pnpm db:migrate
pnpm build
pnpm start

1. Fork the repository
2. Create a feature branch: git checkout -b feature/my-feature
3. Commit your changes: git commit -m "feat: add my feature"
4. Push to the branch: git push origin feature/my-feature
5. Open a Pull Request

• Run pnpm typecheck before committing
• Run pnpm lint to check code style
• Follow Conventional Commits for commit messages

• AI Digital Human — talking-head video generation with digital avatars
• Manga Drama — comic panel animation with expression switching and camera effects
• Multi-language voiceover — expand TTS voice catalog with more languages