AgentLoom
Multi-Modal-AI-Orchestrator (Reset version),AI Full-modal Full-agent:Text → Image → Music → Lights → Video, Includes "Scenario Director, Parent-Child Theater, Biofeedback, Party Collaboration Station, Scenario Recipes" - five ready-to-use scenario packages. AI全模态+全agent:含 “情境导演、亲子剧场、生物反馈、派对协作台、情境食谱” 五大开箱即用场景
Install / Use
/learn @qq749812679/AgentLoomREADME
🌟 AgentLoom|灵构织机
One loom to orchestrate multi‑modal, multi‑agent creation
简体中文 | English
New: Real-time Text → Image → Music → Lights → Video with true device control (Hue/WLED & GPIO/USB on Raspberry Pi/Arduino). Five scene templates: Director, Kids Theater, Wellness, Party DJ/VJ, Home Recipes. Quick start below.
Why it stands out
- 🧠 Agentic orchestration: LangGraph‑based multi‑agent planning with memory, feedback, and safety.
- 🎛️ Text‑to‑Everything pipeline: one workflow to generate image, compose music, choreograph lights, and render video.
- 💡 Device‑native: first‑class Philips Hue / WLED support to run your actual room.
- 🔌 Pluggable backends: SD WebUI / HTTP models with latency/cost profiling and health checks.
- 🧪 Feedback → Auto‑tuning: collect ratings, learn preferences, and surface optimization suggestions.
- 📦 Task queue UX: pause / resume / retry with persistent progress cards.
- 🧭 First‑run wizard: write
.env+ device params in minutes. - 🏢 Enterprise‑ready: license gating, audit logs, role‑based hooks.
What you can build in 60 seconds
- “Cozy jazz bar” → cinematic image + mellow soundtrack + warm lighting show + MP4
- Upload a photo → matching soundtrack + mood‑aligned lighting
- Drop a song → beat‑synced lighting with presets for Hue/WLED
Quick start
Windows (PowerShell)
cd multiscenpowershell -ExecutionPolicy Bypass -File setup.ps1streamlit run app.py
macOS/Linux
cd multiscen && bash setup.shstreamlit run app.py
Optional frontend (React)
cd multiscen/frontend && npm install && npm start
CLI
mscen gen "sunset jazz bar" --videomscen list
Deployment
- System Requirements
- Python 3.9+; Node 18+ (for React frontend); FFmpeg (video)
- Recommended: 8GB RAM+, GPU optional; Raspberry Pi/Arduino for GPIO/USB
- Environment Variables (.env)
- Minimal
OPENAI_API_KEY=...
- Devices
HUE_BRIDGE_IP=192.168.1.100HUE_USERNAME=your_hue_usernameWLED_IP=192.168.1.101GPIO_ENABLED=trueARDUINO_SERIAL=COM3(or/dev/ttyUSB0)
- Minimal
- Example .env
OPENAI_API_KEY=sk-...
HUE_BRIDGE_IP=192.168.1.100
HUE_USERNAME=your_hue_username
WLED_IP=192.168.1.101
GPIO_ENABLED=true
ARDUINO_SERIAL=COM3
- Docker (optional)
docker-compose up -d
Architecture at a glance
flowchart LR
subgraph Frontend [React / Tailwind]
UI[Studio / Templates / Devices]
end
subgraph Backend [Streamlit + FastAPI]
API[REST/WebSocket]
Orchestrator[LangGraph Orchestrator]
Memory[Conversation/Preference]
Safety[Safety Filters]
end
subgraph Connectors
IMG[Image Backends]
MUS[Music Backends]
TTS[STT/TTS]
HUE[Philips Hue]
WLED[WLED]
GPIO[GPIO/USB]
end
UI --> API
API --> Orchestrator
Orchestrator --> Memory
Orchestrator --> Safety
Orchestrator --> IMG
Orchestrator --> MUS
Orchestrator --> TTS
Orchestrator --> HUE
Orchestrator --> WLED
Orchestrator --> GPIO
sequenceDiagram
participant User
participant UI as Frontend UI
participant API as FastAPI
participant Orc as LangGraph Orchestrator
participant IMG as Image Backend
participant MUS as Music Backend
participant HUE as Hue/WLED/GPIO
participant VID as Video Composer
User->>UI: Prompt / select template
UI->>API: Create job (text-to-all)
API->>Orc: Plan agents & timeline
Orc->>IMG: Generate image(s)
Orc->>MUS: Compose music / beat
Orc->>HUE: Program lights (beat-synced)
Orc->>VID: Render video
Orc->>API: Results (urls/preview)
API->>UI: Live updates & final assets
- Backend: Streamlit + FastAPI + LangGraph (feedback learner, model optimizer)
- Frontend: React + TypeScript + Tailwind (real‑time agent/queue views)
- Connectors: SD WebUI, HTTP image/music/STT/TTS, Hue/WLED devices
- Enterprise: License server, audit logger, security hooks
Live demos
- Web UI: Text‑to‑Everything studio, real‑time device control
- Mock API:
docker-compose up -d mock-api - Short video/GIFs: coming soon (PRs welcome!)
Roadmap (highlights)
- Strategy‑based routing (lowest latency / lowest cost / balanced)
- Auto‑tuned parameter replay (one‑click “apply suggestions”)
- Multi‑user spaces and approval workflows
Contributing
- Star ⭐ if this saves your time; PRs welcome
- Use Issues for bugs/ideas, Discussions for design talks
- Sponsors/Backers: see
.github/FUNDING.yml
License
MIT — build amazing things.
