From creative brief to production package in minutes. Powered by Gemini.
MUSE is an AI Creative Director that transforms a creative brief into a complete production package:
| Stage | Output | Model |
|---|---|---|
| 📝 Script | Structured screenplay (scenes, dialogue, camera notes) | Gemini 2.5 Flash |
| 🎨 Style Bible | Visual direction (colors, lighting, camera, film references) | Gemini 2.5 Flash |
| 🖼️ Storyboard | Frame-by-frame breakdown with composition notes | Gemini 2.5 Flash |
| 🔍 Review | AI creative critique with approval/revision cycle | Gemini 2.5 Flash |
| 🖌️ Images | Generated storyboard frames matching the style bible | Gemini 3.1 Flash Image |
User Brief (text)
→ Stage 1: Script Generator (Gemini 2.5 Flash, JSON output)
→ Stage 2: Style Bible Compiler (inherits script context)
→ Stage 3: Storyboard Artist (inherits script + style)
→ Stage 4: Creative Review (critique-correct-verify loop)
→ Stage 5: Image Generator (Gemini 3.1 Flash Image)
→ Complete Production Package
Key Design Decisions:
- Structured JSON contracts between pipeline stages ensure reliable data flow
- Global Lock style string from the style bible is appended to every image prompt for visual consistency
- Critique-correct-verify loop catches quality issues before image generation
- Separate models for text (2.5 Flash) and images (3.1 Flash Image) for optimal quality
- Frontend: Next.js 15 + TypeScript + Tailwind CSS
- Backend: Next.js API Routes + Google GenAI SDK (
@google/genai) - Text Model:
gemini-2.5-flashwithresponseMimeType: "application/json" - Image Model:
gemini-3.1-flash-image-previewwithresponseModalities: ["IMAGE", "TEXT"] - Deployment: Vercel (serverless)
- Node.js 20+
- Google AI API key (get one here)
git clone https://github.com/grepautomation/muse-agent.git
cd muse-agent
npm install
echo "GOOGLE_API_KEY=your_key_here" > .env.local
npm run devOpen http://localhost:3000.
npx vercel --prod
npx vercel env add GOOGLE_API_KEY productionBrief: "A 30-second cyberpunk short: a street vendor sells memories in neon-lit alleyways of Neo-Tokyo. A customer buys their dead lover's last memory."
Result:
- 📝 Script: "The Memory Vendor" — 3 scenes, 11 shots, 2 characters
- 🎨 Style Bible: Neon-noir palette (#FF00FF, #00FFFF, #1A1A2E), anamorphic 2.39:1, Blade Runner meets Wong Kar-wai
- 🖼️ Storyboard: 11 frames with detailed composition and camera notes
- 🔍 Review: Grade A, approved in 1 round
- 🖌️ Images: 11 generated storyboard frames, all matching the style bible
Built by Bastion Harbor Studios — AI-native filmmaking since 2021, 250+ shipped projects, Top 50 AWS AI Artists.
MIT