I recommend using vanilla A-Frame and aframe-state-component with static templating over aframe-react. React wastes a lot of cycles and incurs a lot of memory garbage. aframe-react is often abused where it is too easy to place 3D/real-time logic at the React layer, causing poor performance (e.g., doing React renders on ticks). aframe-react applications frequently ignore the prescribed ECS framework of A-Frame. Internally, React does tons of computation to compute what changed, and flushes it to the entire application. It is apparent React ecosystem does not care much about memory as most examples allocate functions and objects in the render method, and where immutables are popular. With only ~10ms per frame to do all computation, there is little room for React's massive system.

Install with npm or yarn.

npm install --save aframe aframe-react react react-dom
yarn add aframe aframe-react react react-dom

import 'aframe';
import 'aframe-particle-system-component';
import {Entity, Scene} from 'aframe-react';
import React from 'react';
import ReactDOM from 'react-dom';

class VRScene extends React.Component {
  render () {
    return (
      <Scene>
        <Entity geometry={{primitive: 'box'}} material={{color: 'red'}} position={{x: 0, y: 0, z: -5}}/>
        <Entity particle-system={{preset: 'snow'}}/>
        <Entity light={{type: 'point'}}/>
        <Entity gltf-model={{src: 'virtualcity.gltf'}}/>
        <Entity text={{value: 'Hello, WebVR!'}}/>
      </Scene>
    );
  }
}

ReactDOM.render(<VRScene/>, document.querySelector('#sceneContainer'));

See aframe-react-boilerplate for a basic example.

A-Frame is a web framework for building virtual reality experiences. Since A-Frame is built on top of the DOM, web libraries such as React, Vue.js, Angular, Ember.js, d3.js are able to sit cleanly on top of A-Frame.

A-Frame is an entity-component-system (ECS) framework exposed through HTML. ECS is a pattern used in game development that favors composability over inheritance, which is more naturally suited to 3D scenes where objects are built of complex appearance, behavior, and functionality. In A-Frame, HTML attributes map to components which are composable modules that are plugged into <a-entity>s to attach appearance, behavior, and functionality.

Released on the same day as A-Frame, aframe-react is a very thin layer on top of A-Frame to bridge with React. aframe-react passes React props to directly A-Frame using refs and .setAttribute(), bypassing the DOM. This works since A-Frame's .setAttribute()s are able to take non-string data such as objects, arrays, or elements and synchronously modify underlying 3D scene graph.

// aframe-react's <Entity/> React Component
<Entity geometry={{primitive: 'box', width: 5}} position="0 0 -5"/>

// renders
<a-entity>

// and then applies the data directly to the underlying 3D scene graph, bypassing the DOM.
<a-entity>.setAttribute('geometry', {primitive: 'box', width: 5});
<a-entity>.setAttribute('position', '0 0 -5');

aframe-react provides the best of both worlds between A-Frame and React, the 3D and VR-oriented entity-component architecture of A-Frame, and the view and state management ergonomics of React, without penalties of attempting to use React for a VR application.

A-Frame and aframe-react gives some viability for React to use in VR applications. The value proposition of React is limited to the 2D Web:

• Improve rendering performance for 2D web pages by reducing calls to the browser's 2D layout engine via the virtual DOM.
• Provide an predictable application structure for large 2D web applications through a structured nesting of React components, data binding, and one-way data flow.

However, these propositions fall short in 3D and VR applications:

• In 3D and VR applications, including A-Frame, the 2D browser layout engine is not touched. Thus React's reconciliation engine can become an unpredictable performance liability for little benefit.
• In 3D and VR applications, objects are not structured in a top-down hierarchy like 2D web applications. 3D objects can exist anywhere and interact with any other 3D object in a many-to-many manner. But React's paradigm prescribes data flow from parent-to-child, which makes it restrictive for 3D and VR applications.

A-Frame and aframe-react gives meaning and purpose to React: A-Frame provides an actual DOM for React to reconcile, diff, and bind to.

React also raises questions around performance; React is a very large library designed for a 60fps 2D Web, is it equipped to handle 90fps VR applications? 90 prop updates per second per object for many objects through React's engine causes concern in the overhead added for each operation.

aframe-react lets A-Frame handle the heavy lifting 3D and VR rendering and behavior. A-Frame is optimized from the ground up for WebVR with a 3D-oriented entity-component architecture. And aframe-react lets React focus on what it's good at: views, state management, and data binding.

A-Frame's entity-component-system (ECS) pattern is tailored for 3D and VR applications. What React did for 2D web applications is what ECS did for 3D applications in terms of providing a useful abstraction. ECS promotes composability over hierarchy. In 3D applications, composability refers to composition of appearance, behavior, and logic rather than having fixed blocks. This lets us do things like define a flying robot that explodes on contact and makes robot sounds by snapping together a pre-existing model component, explode component, event handler component, sound component, and flying component.

Unfortunately, React (and the 2D web for that matter) is heavily hierarchical, which is not suited for 3D. Whereas 2D web development consisted of structuring and laying out from an assorted set of fixed 2D elements (e.g., <p>, <a>, <img>, <form>), 3D development involves objects that are infinite in complexity and type. ECS provides an easy way of defining those objects by mixing and matching plug-and-play components.

With aframe-react, we get the best of both worlds. The 3D and VR architecture of A-Frame, and the view and state ergonomics of React. React can be used to bind application and state data to the values of A-Frame components. And we still have access to all the features and performance of A-Frame as well as A-Frame's community component ecosystem.

aframe-react's API is very thin on top of A-Frame, less than 200 lines of source code. The API consists of just two React Components: <Entity/> and <Scene/>.

<Entity/> wraps <a-entity>, the entity piece of the entity-component-system pattern. Plug in A-Frame components as React props to attach appearance, behavior, or functionality to the <Entity/>.

<Scene>
  <Entity
    geometry={{primitive: 'box', width: 5}}
    material={{color: red, roughness: 0.5, src: texture.png}}
    scale={{x: 2, y: 2, z: 2}}
    position={{x: 0, y: 0, z: -5}}/>
</Scene>

Community A-Frame components can be imported and installed through npm:

import 'aframe-particle-system-component';
import 'aframe-mountain-component';

// ...

<Scene>
  <Entity mountain/>
  <Entity particle-system={{preset: 'snow', particleCount: 5000}}/>
</Scene>

To use A-Frame primitives, provide the primitive prop with the primitive's element name (e.g., a-sphere). Mappings can be applied the same as in HTML through React props:

<Entity primitive='a-box' color="red" position="0 0 -5"/>
<Entity primitive='a-sphere' color="green" position="-2 0 -3"/>
<Entity primitive='a-cylinder' color="blue" position="2 0 -3"/>
<Entity primitive='a-sky' src="sechelt.jpg"/>

To register event handlers, use the events prop. events takes a mapping of event names to event handler(s). Multiple event handlers can be provided for a single event name by providing an array of functions. Try not to pass in inline functions to not trigger unnecessary React renders. Pass in binded functions instead.

For example, using the synthetic click event provided by A-Frame's cursor component, or a collided event possibly provided by a physics component.

handleClick = () => {
  console.log('Clicked!');
}

handleCollide = () => {
  console.log('Collided!');
}

render() {
  return (
    <Scene>
      <Entity events={{
        click: this.handleClick,
        collided: [this.handleCollide]}}/>
    </Scene>
  );
}

aframe-react does not support React-style onXXX event handlers (e.g., onClick). Unlike 2D web pages, VR sites are composed entirely of custom synthetic event names (which could have hyphens, be all lowercase, be camelCase, all uppercase, camel case, etc.,). The possible event names are infinite. The events prop makes it explicit what the event names to handle are.

Use aframe-react's _ref prop to add a callback to the underlying <a-entity> DOM node:

<Entity _ref={this.entityCallback}/>

<Scene/> extends <Entity/> and renders <a-scene> instead of <a-entity>. Place all <Entity/>s as a child of the <Scene/> Component. There should only be one <Scene/> per page:

<Scene>
  <Entity/>
</Scene>

Let A-Frame and three.js handle the heavy lifting of 3D rendering and behavior, and delegate React to what it was meant for: views, state management, and data binding.

For example, don't create requestAnimationFrames to continuously throw React prop updates to A-Frame. Instead, create an A-Frame component with a tick handler such that everything is done in memory with little overhead. Do however use React to set up initial state and data binding that might configure that A-Frame component.

aframe-react works with Preact. Since aframe-react uses React.Component, we have to tell Webpack to alias that to Preact.Component:

resolve: {
  alias: {
    react: 'preact'
  }
}