Concept-3D is an AI-powered spatial generation and discovery platform. It transforms unstructured text descriptions and 2D concept images into interactive, high-fidelity 3D meshes using TripoSR deep learning models, Chroma vector database semantic retrieval, and a WebGL Three.js interactive viewport.
- Live Web Client: https://concept-3d.vercel.app/
- Repository: https://github.com/GuruMachanica/Concept-3D
- Milestone: Sankalp 2026 National Summit on Innovation & Skills Qualifier (MNNIT Prayagraj)
- Prompt-to-3D Mesh Synthesis: Deep generative reconstruction pipeline converting text prompts and single-view 2D concept images into fully textured 3D geometric meshes (.glb, .obj, .stl).
- TripoSR ML Pipeline: Optimized neural network feed-forward architecture generating production-ready 3D meshes in under 2.4 seconds.
- Vector Semantic Retrieval (ChromaDB): High-dimensional vector indexing allowing context-aware asset search, category clustering, and real-time conceptual similarity ranking.
- Interactive WebGL 3D Studio: Client-side Three.js orbit viewport with real-time wireframe inspection, vertex telemetry, dynamic lighting controls, and multi-angle asset inspection.
- Context-Aware Design Assistant: Embedded AI chat engine with fallback provider routing for design critique, mesh optimization recommendations, and semantic prompt refinement.
- Asset Lifecycle & Review Engine: Relational SQLite metadata store with user feedback rating queues and automated GLB export caching.
+-----------------------------------------------------------------------------------+
| CONCEPT-3D PLATFORM |
+-----------------------------------------------------------------------------------+
|
+-----------------------+-----------------------+
| |
v v
+---------------------+ +---------------------+
| React 18 Studio | | FastAPI Backend |
| (Three.js WebGL UI) |<--- REST & WebSockets --->| (Service Engine) |
+---------------------+ +---------------------+
| |
v +-- Intent & NLP Parser
+---------------------+ +-- Chroma Vector Database
| 3D Orbit Viewport | +-- TripoSR Neural ML Engine
| Wireframe & Shading| +-- GLB Export Pipeline
| Spatial Telemetry | +-- Design Assistant Chat
+---------------------+ +-- Asset Review Store
|
v
+---------------------+
| ChromaDB / SQLite |
| Cached 3D Artifacts |
+---------------------+
sequenceDiagram
autonumber
actor User as 3D Designer
participant UI as React Three.js Studio
participant API as FastAPI Router (Port 8011)
participant NLP as Intent Parser
participant Chroma as Chroma Vector Index
participant ML as TripoSR Generative Engine
participant Cache as 3D Asset Cache Store
User->>UI: Submit Concept Text / Image
UI->>API: POST /api/generate_from_image_async
API->>NLP: Extract Semantic Spatial Tokens
par Semantic Similarity Query
NLP->>Chroma: Vector Nearest Neighbor Query
Chroma-->>API: Pre-Indexed Related Assets
and Neural Mesh Generation
API->>ML: Pass Normalized Concept Tensor
ML->>ML: Run 3D TripoSR Reconstruction (2.1s)
ML-->>Cache: Export Standard GLB Mesh Artifact
end
API-->>UI: Return Job Complete & Asset URI
UI->>UI: Render Interactive 3D Mesh in Three.js Canvas (60 FPS)
| Metric | Target Specification | Achieved Benchmark |
|---|---|---|
| Generation Latency (GPU - CUDA) | < 3.0s |
2.14s |
| Generation Latency (CPU fallback) | < 15.0s |
9.80s |
| Vector Search Query Latency | < 50ms |
18.5ms |
| Target Polygon Density | 15,000 - 80,000 vertices |
Adaptive LOD |
| Supported Export Formats | GLB, GLTF, OBJ, STL | Native Binary Export |
| Frontend WebGL Render Rate | 60 FPS |
Hardware-Accelerated |
| Method | Endpoint | Description |
|---|---|---|
POST |
/api/intent |
Analyzes text prompts to extract 3D category & styling intent |
POST |
/api/search |
Performs vector similarity & keyword search across ChromaDB |
POST |
/api/chat |
Context-aware AI design assistant conversation |
POST |
/api/generate_from_image |
Synchronous image-to-3D mesh generation |
POST |
/api/generate_from_image_async |
Queues asynchronous 3D generation job |
GET |
/api/generate_status/{job_id} |
Polls progress and retrieves completed 3D asset URL |
POST |
/api/reviews/submit |
Submits model rating and architectural feedback |
GET |
/api/reviews/{model_id} |
Retrieves user reviews and community ratings |
Interactive Swagger Documentation: http://127.0.0.1:8011/docs
- Python 3.10+ (Python 3.11/3.12 recommended)
- Node.js 18+ & npm
- Git
# On Windows (PowerShell):
.\scripts\setup_venv.ps1
# On Linux / macOS:
./scripts/setup_venv.shCopy Backend/.env.example to Backend/.env and configure keys:
GROQ_API_KEY=your_groq_api_key_here
PORT=8011In Terminal 1 (Backend on Port 8011):
.venv\Scripts\python.exe -m uvicorn Backend.main:app --host 0.0.0.0 --port 8011 --reloadIn Terminal 2 (Frontend on Port 5173):
cd Frontend
npm install
npm run devOpen http://localhost:5173 in your browser.
- Sandboxed Mesh Generation: Model synthesis occurs within isolated temp directories with automated garbage collection.
- Zero Secret Leakage: All AI provider API keys are loaded strictly via
.envand guarded by.gitignore. - CORS Protection: REST and WebSocket connections are bounded to authorized studio endpoints.
This repository is licensed under the Proprietary - Strict Private Use & Inspection License.
See the LICENSE file for terms and restrictions.
Copyright (c) 2026 Mohammad Huzaifa & Contributors. All rights reserved.