@nexus-render/core

High-performance headless 3D rendering for Node.js — written in Rust with wgpu, exposed to TypeScript via napi-rs.

Render 3D scenes entirely on the server: no browser, no display, no DOM.

Features

• 100% headless — runs in Node.js, Docker, or cloud servers with no display required
• GPU accelerated — uses native GPU via wgpu (Vulkan / Metal / DX12) with automatic software fallback
• GLTF/GLB loading — load real 3D assets from disk
• Built-in primitives — cube and sphere out of the box
• Multiple output formats — raw RGBA Uint8Array, JPEG buffer, depth map Float32Array
• Three.js compatible — right-handed Y-up coordinate system, [x, y, z, w] quaternions

Requirements

• Node.js >= 18
• Linux x86_64 (glibc) — additional platforms planned

Installation

npm install @nexus-render/core

Quick Start

import { RenderEngine } from '@nexus-render/core'

const engine = new RenderEngine({ width: 1280, height: 720, enableGpu: false })

// Build a scene
const cubeId = engine.addPrimitive('cube')
engine.setTransform(cubeId, [0, 0, 0], [0, 0, 0, 1])
engine.addDirectionalLight([0, -1, -0.5], 0.8)
engine.setCamera([0, 2, 5], [0, 0, 0], 60)

// Render to raw RGBA
const pixels = engine.renderRaw('default')
// → Uint8Array of length width * height * 4

// Render to JPEG
const jpeg = engine.renderFrameJpeg('default', 90)
// → Uint8Array starting with JPEG magic bytes FF D8 FF

// Render depth map
const depth = engine.renderDepth('default')
// → Float32Array of length width * height, values in [0.0, 1.0]

// Load a GLTF/GLB model
const modelId = engine.loadModel('/path/to/model.glb')
engine.setTransform(modelId, [1, 0, 0], [0, 0, 0, 1])

API

new RenderEngine(options)

| Option | Type | Description | |---|---|---| | width | number | Render target width in pixels | | height | number | Render target height in pixels | | enableGpu | boolean | Request a physical GPU; falls back to software if unavailable |

Methods

| Method | Returns | Description | |---|---|---| | addPrimitive(type) | number | Add a built-in primitive ("cube" or "sphere"), returns a unique entity ID | | setTransform(id, position, rotation) | void | Set world-space position [x,y,z] and quaternion [x,y,z,w] | | setCamera(position, target, fovDegrees) | void | Configure the virtual camera | | addDirectionalLight(direction, intensity) | void | Add a directional light | | loadModel(filePath) | number | Load a GLTF/GLB file, returns a unique entity ID | | renderRaw(cameraId) | Uint8Array | Raw RGBA pixel buffer — width × height × 4 bytes | | renderFrameJpeg(cameraId, quality) | Uint8Array | JPEG-encoded frame — quality range 1–100 | | renderDepth(cameraId) | Float32Array | Depth map — width × height values in [0.0, 1.0] |

Coordinate system

Matches Three.js conventions:

• Right-handed, Y-up
• +X = right, +Y = up, +Z = toward the viewer
• Quaternions in [x, y, z, w] order

License

MIT