🎨 Forge Master

AI-powered 3D model generation from text prompts

Forge Master is a production-ready AI-powered platform that transforms text descriptions into high-quality 3D models. Simply describe what you want, and our AI agents will generate, optimize, and deliver production-ready 3D assets in multiple formats.

🚀 Try It Live

Production URL: https://forge-master-frontend-525900378413.europe-west1.run.app

✨ Features

• Text-to-3D: Describe your model in plain English
• AI-Powered Quality: 4 intelligent agents ensure production-ready results
• Multiple Formats: Download in GLB, OBJ, FBX, and STL
• Interactive Viewer: Inspect your model in 3D before downloading
• Fast Generation: 70-140 seconds from prompt to download
• Quality Guaranteed: AI quality assessment with scores and analysis

🔒 Security Status

Backend APIs are now protected! The GPU service (most expensive component) uses service-to-service authentication:

• ✅ GPU Service: Protected - Only accessible by agent-service (not public)
• ✅ Agent Service: Public - Accessible via web frontend only
• ✅ Frontend: Public - Anyone can use the web interface

What this means:

• Users can access the web interface normally
• Direct API abuse is prevented (GPU endpoint is protected)
• Cost risk from the expensive GPU operations is mitigated

🎯 Quick Start

For Users

1. Visit https://forge-master-frontend-525900378413.europe-west1.run.app
2. Enter a text prompt (e.g., "A futuristic sports car")
3. Click "Generate 3D Model"
4. Wait ~90 seconds
5. Download your model in your preferred format

Note: This is a demo deployment. Please use responsibly.

🏗️ Architecture

┌─────────────┐
│   Browser   │  Next.js Frontend + 3D Viewer
└──────┬──────┘
       │
       ▼
┌─────────────┐
│Agent Service│  4 AI Agents (Gemini)
└──────┬──────┘  • Prompt Enhancement
       │         • Generation Coordination
       │         • Quality Assessment
       │         • Iterative Improvement
       ▼
┌─────────────┐
│ GPU Service │  Imagen 4 + InstantMesh (L4 GPU)
└──────┬──────┘  • Text → Image
       │         • Image → Multi-view
       │         • Multi-view → 3D Mesh
       ▼
┌─────────────┐
│Cloud Storage│  GLB, OBJ, FBX, STL files
└─────────────┘

🛠️ Tech Stack

Frontend

• Next.js 15 - Modern React framework
• shadcn/ui - Polished UI components
• React Three Fiber - 3D rendering and viewer
• Tailwind CSS - Styling

Backend

• FastAPI - High-performance Python API
• Imagen 4 - Text-to-image generation
• InstantMesh - Image-to-3D conversion
• Gemini AI - Prompt enhancement & quality assessment

Infrastructure

• Google Cloud Run - Serverless deployment
• NVIDIA L4 GPU - GPU acceleration
• Cloud Storage - Model file hosting

📖 Documentation

• Quick Start Guide - Get started in 30 seconds
• Project Summary - Complete overview and features
• Deployment Guide - Redeployment and configuration
• GPU Service README - GPU service details
• Agent Service README - Agent pipeline details
• Frontend README - Frontend implementation

💡 Example Prompts

Try these prompts to see what Forge Master can create:

A wooden chair with curved armrests
A medieval fantasy sword with ornate handle
A futuristic sports car with sleek design
A modern minimalist coffee table
A vintage rotary telephone

📊 Performance

• Generation Time: 70-140 seconds
• Success Rate: >95%
• Average Quality Score: 8.2/10
• Supported Formats: GLB, OBJ, FBX, STL

🎨 Quality System

Each generated model receives an AI quality score:

• 🟢 8.5-10: Production Ready ✅
• 🟡 7.0-8.5: Good Quality
• 🔴 <7.0: Needs Improvement

Models that score below the target are automatically improved through iterative refinement.

💰 Cost (Self-Hosting)

Approximate cost per generation:

• GPU time: ~$0.50
• Imagen API: ~$0.04
• Gemini API: ~$0.001
• Storage: ~$0.001

Total: ~$0.56 per model

🔗 Service URLs

Service	URL	Access	Purpose
Frontend	https://forge-master-frontend-525900378413.europe-west1.run.app	Public	Web interface
Agent Service	https://forge-master-agents-7n53rqwksq-ew.a.run.app	Public (Frontend only)	AI orchestration
GPU Service	https://forge-master-gpu-7n53rqwksq-ew.a.run.app	Protected (IAM)	3D generation

Note: The GPU service URL is listed for reference only. Direct access is blocked - it can only be accessed by the agent-service with proper authentication.

📁 Project Structure

cloudrun/
├── frontend/              # Next.js web application
├── agent-service/         # 4-agent AI pipeline
├── gpu-service/           # GPU-accelerated 3D generation
├── README.md             # This file
├── QUICKSTART.md         # Quick start guide
├── PROJECT_SUMMARY.md    # Detailed project overview
└── DEPLOYMENT.md         # Deployment instructions

🚀 Deployment

All services are currently deployed and operational. See DEPLOYMENT.md for redeployment instructions if needed.

Prerequisites:

• Google Cloud Project
• Billing enabled
• Cloud Run, Cloud Build, Cloud Storage APIs enabled
• GOOGLE_API_KEY for Imagen and Gemini

🔒 Security Implementation

Current Security Setup:

✅ Service-to-Service Authentication (Implemented)

• GPU service is protected with Cloud Run IAM authentication
• Only the agent-service can invoke the GPU service
• Direct API calls to GPU endpoint are blocked with 403 Forbidden

🤝 Contributing

This is a hackathon project currently in production. Feel free to:

• Report issues or bugs
• Suggest features or improvements
• Fork and experiment with your own modifications

🔌 Third-Party Integrations & Licenses

This project integrates the following third-party SDKs, APIs, and open-source libraries, all used in compliance with their respective terms and licenses:

Google Cloud Services (Proprietary)

• Google Imagen 4 (via Imagen 4.0 API) - Text-to-image generation (Authorized with GOOGLE_API_KEY)
• Google Gemini AI - Prompt enhancement and quality assessment (Authorized with GOOGLE_API_KEY)
• Google Cloud Run - Serverless deployment platform
• Google Cloud Storage - Object storage for 3D model files

Open Source AI Models

• InstantMesh by TencentARC - Image-to-3D mesh conversion
  • License: Apache 2.0
  • Components: Zero123++ (multi-view synthesis), LRM (Large Reconstruction Model)

Frontend Libraries (Open Source)

• Next.js - React framework (MIT License)
• React - UI library (MIT License)
• React Three Fiber - 3D rendering (MIT License)
• Three.js - 3D graphics library (MIT License)
• shadcn/ui - UI component library (MIT License)
• Tailwind CSS - CSS framework (MIT License)
• Radix UI - Headless UI primitives (MIT License)

Backend Libraries (Open Source)

• FastAPI - Python web framework (MIT License)
• PyTorch - Deep learning framework (BSD-3-Clause License)
• Hugging Face Transformers - NLP models (Apache 2.0)
• Hugging Face Diffusers - Diffusion models (Apache 2.0)

3D Processing Libraries (Open Source)

• Trimesh - Python 3D mesh library (MIT License)
• PyMeshLab - MeshLab Python bindings (GPL License)
• PyMCubes - Marching cubes implementation (BSD License)
• xatlas - UV unwrapping library (MIT License)

Additional Dependencies

• OpenCV - Computer vision library (Apache 2.0)
• Pillow - Image processing (HPND License)
• NumPy - Numerical computing (BSD License)
• SciPy - Scientific computing (BSD License)

Compliance Statement: All third-party integrations are used in accordance with their respective licensing terms. Google Cloud APIs are accessed with valid API credentials. All open-source software is properly attributed and used under permissive licenses (MIT, Apache 2.0, BSD) or with proper compliance (GPL).

📝 License

This project was built as a hackathon submission. The original code is available under MIT License, while third-party components retain their respective licenses as listed above.

🎉 Credits

Built with:

• Google Cloud Run - Serverless platform
• Imagen 4 - Text-to-image AI
• InstantMesh - Image-to-3D conversion
• Gemini AI - Prompt enhancement & QA
• Next.js - React framework
• shadcn/ui - UI components
• React Three Fiber - 3D rendering

---

Live Demo: https://forge-master-frontend-525900378413.europe-west1.run.app