---

name: nepa-viewport-3d description: AI-powered 3D modeling and VR preview workspace for the NEPA AI platform. Handles text-to-3D generation (Shap-E), model optimization, VR export, and 3D asset management. Use when generating 3D models, optimizing meshes, or preparing VR-ready assets in NEPA AI.

3D Viewport Workspace

805 lines — 13 methods. Text-to-3D with VR-ready export pipeline.

Prerequisites

pip install torch numpy pillow trimesh replicate requests scipy
pip install shap-e @ git+https://github.com/openai/shap-e.git

C++ acceleration: libomni_3d_modeling.a (NEPA Omni 3D) — included in the AXON product ZIP. Link with: gcc -o app app.c -L./libs -lomni_3d_modeling

Source

/home/billk/projects/nepa-ai-monorepo.BAK/vscode_forked/nepa-ai-backend/viewport_3d_workspace.py

Setup

cd ~/projects/nepa-ai-monorepo.BAK/vscode_forked/nepa-ai-backend
python run_nepa.py

Key Methods

| Method | Description | |--------|-------------| | generate_3d_from_text(prompt) | Text-to-3D via Shap-E | | generate_3d_from_image(image) | Image-to-3D reconstruction | | optimize_model(path, target_polys) | Polygon reduction / LOD generation | | generate_texture(model, prompt) | AI texture generation for mesh | | generate_collision_mesh(model) | Collision mesh for physics | | convert_format(path, format) | OBJ ↔ glTF ↔ FBX ↔ STL ↔ USDZ | | prepare_vr_export(model) | VR/AR optimization + USDZ packaging | | generate_lod_chain(model) | LOD0-LOD3 chain generation | | preview_3d(model) | Three.js web preview | | batch_process(models, ops) | Batch model processing |

Example Usage

import asyncio
from viewport_3d_workspace import Viewport3DWorkspace

async def main():
    ws = Viewport3DWorkspace()
    
    # Generate a 3D model
    model = await ws.generate_3d_from_text("sleek futuristic helmet, sci-fi, metallic")
    
    # Optimize for real-time use
    optimized = await ws.optimize_model(model["path"], target_polys=5000)
    
    # Prepare for VR export
    vr_ready = await ws.prepare_vr_export(optimized["path"])
    print(f"USDZ: {vr_ready['usdz_path']}")
    
    # Generate from image
    from_img = await ws.generate_3d_from_image("product_photo.jpg")

asyncio.run(main())

API Endpoints

POST /api/3d/generate-from-text
POST /api/3d/generate-from-image
POST /api/3d/optimize
POST /api/3d/texture
POST /api/3d/convert
POST /api/3d/vr-export
POST /api/3d/lod-chain

AXON Store

Available at: https://axon.nepa-ai.com (viewport-3d-workspace, $97) Download: /downloads/viewport-3d-workspace.zip

LiDAR Capture — No Paywall Pipeline

iPhone 16 Pro setup:

1. Install Record3D (free tier) from App Store
2. Capture scan → export as .r3d format
3. AirDrop or Telegram the file to machine
4. NEPA processes via libomni_3d_modeling.a + libomni_vr_ar.a

Supported input formats: .r3d, .ply, .usdz, .obj, .las Output: processed point cloud, mesh, or VR-ready asset

Note: Raw depth frames from Record3D free tier are enough for the omni C++ pipeline. No Polycam/Scaniverse subscription needed.

---

Full version: https://axon.nepa-ai.com/products