These demos are real, you can click them! They contain the full code, too.
npm install three react-three-fiber
React-three-fiber is a small React renderer for Threejs. Why, you might ask? React was made to drive complex tree structures, it makes just as much sense for Threejs as for the DOM. Building a dynamic scene graph becomes so much easier because you can break it up into declarative, re-usable components with clean, reactive semantics. This also opens up the ecosystem, you can now apply generic packages for state, animation, gestures and so on.
This is a small reconciler config with a few additions for interaction and hooks holding it all together. It does not know or care about Three internals, it uses heuristics for objects and attributes, so that we can get away without creating a strong dependency. Three is constantly changing, we don't want to rely on a specific version or chase their release cycle. This library works with version 1 as well as their latest. At the same time we don't want to alter any rules, if something works in Threejs in a specific way, it will be the same here.
Copy the following into a project to get going. Here's the same running in a code sandbox.
import * as THREE from 'three'
import React from 'react'
import ReactDOM from 'react-dom'
import { Canvas } from 'react-three-fiber'
function Thing({ vertices }) {
return (
<group ref={ref => console.log('we have access to the instance')}>
<line>
<geometry
attach="geometry"
vertices={vertices.map(v => new THREE.Vector3(...v))}
onUpdate={self => (self.verticesNeedUpdate = true)}
/>
<lineBasicMaterial attach="material" color="black" />
</line>
<mesh
onClick={e => console.log('click')}
onPointerOver={e => console.log('hover')}
onPointerOut={e => console.log('unhover')}>
<octahedronGeometry attach="geometry" />
<meshBasicMaterial attach="material" color="peachpuff" opacity={0.5} transparent />
</mesh>
</group>
)
}
ReactDOM.render(
<Canvas>
<Thing vertices={[[-1, 0, 0], [0, 1, 0], [1, 0, 0], [0, -1, 0], [-1, 0, 0]]} />
</Canvas>,
document.getElementById('root')
)
The Canvas
object is your portal into Threejs. It renders Threejs elements, not DOM elements!
<Canvas
children // Either a function child (which receives state) or regular children
gl // Props that go into the default webGL-renderer
camera // Props that go into the default camera
raycaster // Props that go into the default raycaster
vr = false // Switches renderer to VR mode, then uses gl.setAnimationLoop
orthographic = false // Creates an orthographic camera if true
pixelRatio = undefined // You could provide window.devicePixelRatio if you like
invalidateFrameloop = false // When true it only renders on changes, when false it's a game loop
updateDefaultCamera = true // Adjusts default camera on size changes
onCreated /> // Callback when vdom is ready (you can block first render via promise)
You can give it additional properties like style and className, which will be added to the container (a div) that holds the dom-canvas element.
Canvas will create a translucent webGL-renderer with the following properties:
antialias: true, alpha: true, setClearAlpha(0)
A default perspective camera:
fov: 75, near: 0.1, far: 1000, position.z: 5
Or a default orthographic camera if Canvas.orthographic is true:
near: 0.1, far: 1000, position.z: 5
A default scene (into which all the jsx is rendered) and a raycaster. You do not have to use any of these objects, look under "receipes" down below if you want to bring your own.
You can use Three's entire object catalogue and all properties. When in doubt, always consult the docs.
<mesh
visible
userData={{ test: "hello" }}
position={new THREE.Vector3(1, 2, 3)}
rotation={new THREE.Euler(0, 0, 0)}
geometry={new THREE.SphereGeometry(1, 16, 16)}
material={new THREE.MeshBasicMaterial({ color: new THREE.Color('indianred'), transparent: true })} />
All properties that have a .set()
method (colors, vectors, euler, matrix, etc) can be given a shortcut. For example THREE.Color.set can take a color string, hence instead of color={new THREE.Color('peachpuff')}
you can do color="peachpuff"
. Some set
methods take multiple arguments (vectors for instance), in this case you can pass an array.
You can stow away non-Object3D primitives (geometries, materials, etc) into the render tree so that they become managed and reactive. They take the same properties they normally would, constructor arguments are passed with args
. Using the attach
property objects bind automatically to their parent and are taken off it once they unmount.
The following is the same as above, but it's leaner and critical properties aren't re-instantiated on every render.
<mesh visible userData={{ test: "hello" }} position={[1, 2, 3]} rotation={[0, 0, 0]}>
<sphereGeometry attach="geometry" args={[1, 16, 16]} />
<meshStandardMaterial attach="material" color="indianred" transparent />
</mesh>
You can nest primitive objects—which is good for awaiting async textures and such. You could use React-suspense if you wanted!
<meshBasicMaterial attach="material">
<texture attach="map" format={THREE.RGBFormat} image={img} onUpdate={self => img && (self.needsUpdate = true)} />
</meshBasicMaterial>
Sometimes attaching isn't enough. For example, this code attaches effects to an array called "passes" of the parent effectComposer
. Note the use of attachArray
which adds the object to the target array and takes it out on unmount:
<effectComposer>
<renderPass attachArray="passes" />
<glitchPass attachArray="passes" renderToScreen />
</effectComposer>
You can also attach to named parent properties using attachObject={[target, name]}
, which adds the object and takes it out on unmount. The following adds a buffer-attribute to parent.attributes.position.
<bufferGeometry>
<bufferAttribute
attachObject={['attributes', 'position']}
count={vertices.length / 3}
array={vertices}
itemSize={3} />
</bufferGeometry>
If you want to reach into nested attributes (for instance: mesh.rotation.x
), just use dash-case:
<mesh rotation-x={1} material-color="lightblue" geometry-vertices={newVertices} />
You can use the primitive
placeholder for that. You can still give it properties or attach nodes to it.
const mesh = new THREE.Mesh()
return <primitive object={mesh} position={[0, 0, 0]} />
The apply
function extends three-fibers catalogue of known native elements. These objects become available and can now be directly instantiated.
import { apply } from 'react-three-fiber'
import { EffectComposer } from './postprocessing/EffectComposer'
import { RenderPass } from './postprocessing/RenderPass'
apply({ EffectComposer, RenderPass })
<effectComposer>
<renderPass />
</effectComposer>
THREE objects that implement their own raycast
method (for instance meshes, lines, etc) can be interacted with by declaring events on the object. We support pointer events (you need to polyfill them yourself), clicks and wheel-scroll.
Additionally there's a special onUpdate
that is called every time the object is updated with fresh props (as well as when it's first being created).
The event data you receive contains the browser event as well as the Threejs event data (object, point, distance, etc) all merged into one object.
<mesh
onClick={e => console.log('click')}
onWheel={e => console.log('wheel spins')}
onPointerUp={e => console.log('up')}
onPointerDown={e => console.log('down')}
onPointerOver={e => console.log('hover')}
onPointerOut={e => console.log('unhover')}
onPointerMove={e => console.log('move')}
onUpdate={self => console.log('props have been updated')}
onPointerDown={e => {
// Only the mesh closest to the camera will be processed
e.stopPropagation()
// You may optionally capture the target
e.target.setPointerCapture(e.pointerId)
}}
onPointerUp={e => {
e.stopPropagation()
// Optionally release capture
e.target.releasePointerCapture(e.pointerId)
}}
All hooks can only be used inside the Canvas element because they rely on context updates!
This hooks gives you access to all the basic objects that are kept internally, like the default renderer, scene, camera. It also gives you the size of the canvas in screen and viewport coordinates. When you resize the canvas, or the browser window, your component will be updated with fresh values.
import { useThree } from 'react-three-fiber'
const {
gl, // WebGL renderer
canvas, // canvas the dom element that was created
scene, // Default scene
camera, // Default camera
size, // Bounds of the view (which stretches 100% and auto-adjusts)
viewport, // Bounds of the viewport in 3d units + factor (size/viewport)
aspect, // Aspect ratio (size.width / size.height)
invalidate, // Invalidates a single frame (for <Canvas invalidateFrameloop />)
intersect, // Calls onMouseMove handlers for objects underneath the cursor
setDefaultCamera // Sets the default camera
} = useThree()
If you're running effects, postprocessings, controls, etc that need to get updated every frame, useRender gives you access to the render-loop. You receive the internal state as well, which is the same as what you would get from useThree.
import { useRender } from 'react-three-fiber'
// Subscribes to the render-loop, gets cleaned up automatically when the component unmounts
useRender(state => console.log("I'm in the render-loop"))
// Add a "true" as the 2nd argument and you take over the render-loop completely
useRender(({ gl, scene, camera }) => gl.render(scene, camera), true)
Sometimes objects have to be updated imperatively. You could update the parts that you can access declaratively and then call onUpdate={self => ...}
, or there's useUpdate.
import { useUpdate } from 'react-three-fiber'
const ref = useUpdate(
geometry => {
geometry.addAttribute('position', getCubeVertices(x, y, z))
geometry.attributes.position.needsUpdate = true
geometry.computeBoundingSphere()
},
[x, y, z], // execute only if these properties change
)
return <bufferGeometry ref={ref} />
Materials and such aren't normally re-created for every instance using it. You may want to share and re-use resources. This can be done imperatively simply by maintaining the object yourself, but it can also be done declaratively by using refs. useResource
simply creates a ref and re-renders the component when it becomes available next frame. You can pass this reference on, or even channel it through a context provider.
import { useResource } from 'react-three-fiber'
const [ref, material] = useResource()
return (
<meshBasicMaterial ref={ref} />
{material && (
<mesh material={material} />
<mesh material={material} />
<mesh material={material} />
)}
)
import {
addEffect, // Adds a global callback which is called each frame
invalidate, // Forces view global invalidation
apply, // Extends the native-object catalogue
createPortal, // Creates a portal (it's a React feature for re-parenting)
render, // Internal: Renders three jsx into a scene
unmountComponentAtNode, // Internal: Unmounts root scene
applyProps, // Internal: Sets element properties
} from 'react-three-fiber'
You can use React's built-in memoizing-features (as well as suspense) to build async dependence graphs.
function Image({ url }) {
const texture = useMemo(() => new THREE.TextureLoader().load(url), [url])
return (
<mesh>
<planeBufferGeometry attach="geometry" args={[1, 1]} />
<meshLambertMaterial attach="material" transparent>
<primitive attach="map" object={texture} />
</meshLambertMaterial>
</mesh>
)
}
Managing effects can get quite complex normally. Drop the component below into a scene and you have a live effect. Remove it and everything is as it was without any re-configuration.
import { apply, Canvas, useRender, useThree } from 'react-three-fiber'
import { EffectComposer } from './postprocessing/EffectComposer'
import { RenderPass } from './postprocessing/RenderPass'
import { GlitchPass } from './postprocessing/GlitchPass'
// Makes these objects available as native objects "<renderPass />" and so on
apply({ EffectComposer, RenderPass, GlitchPass })
function Effects({ factor }) {
const { gl, scene, camera, size } = useThree()
const composer = useRef()
useEffect(() => void composer.current.setSize(size.width, size.height), [size])
// This takes over as the main render-loop (when 2nd arg is set to true)
useRender(() => composer.current.render(), true)
return (
<effectComposer ref={composer} args={[gl]}>
<renderPass attachArray="passes" args={[scene, camera]} />
<glitchPass attachArray="passes" factor={factor} renderToScreen />
</effectComposer>
)
}
useRender
allows components to hook into the render-loop, or even to take it over entirely. That makes it possible for one component to render over the content of another. The order of these operations is established by the scene-graph.
function Content({ camera }) {
const scene = useRef()
useRender(({ gl }) => void ((gl.autoClear = true), gl.render(scene.current, camera)), true)
return <scene ref={scene}>{/* ... */}</scene>
}
function HeadsUpDisplay({ camera }) {
const scene = useRef()
useRender(({ gl }) => void ((gl.autoClear = false), gl.clearDepth(), gl.render(scene.current, camera)))
return <scene ref={scene}>{/* ... */}</scene>
}
function Main() {
const camera = useRef()
const { width, height } = useThree().size
return (
<>
<perspectiveCamera
ref={camera}
aspect={width / height}
radius={(width + height) / 4}
onUpdate={self => self.updateProjectionMatrix()}
/>
{camera.current && (
<group>
<Content camera={camera.current} />
<HeadsUpDisplay camera={camera.current} />
</group>
)}
</>
)
}
Stick imperative stuff into useMemo and write out everything else declaratively. This is how you can quickly form reactive, re-usable components that can be bound to a store, graphql, etc.
function Extrusion({ start = [0,0], paths, ...props }) {
const shape = useMemo(() => {
const shape = new THREE.Shape()
shape.moveTo(...start)
paths.forEach(path => shape.bezierCurveTo(...path))
return shape
}, [start, paths])
return (
<mesh>
<extrudeGeometry attach="geometry" args={[shape, props]} />
<meshPhongMaterial attach="material" />
</mesh>
)
}
Then ...
<Extrusion
start={[25, 25]}
paths={[[25, 25, 20, 0, 0, 0], [30, 0, 30, 35,30,35], [30, 55, 10, 77, 25, 95]]}
bevelEnabled
amount={8}
bevelSegments={2}
steps={2}
bevelSize={1}
bevelThickness={1} />
function CrossFade({ url1, url2, disp }) {
const [texture1, texture2, dispTexture] = useMemo(() => {
const loader = new THREE.TextureLoader()
return [loader.load(url1), loader.load(url2), loader.load(disp)]
}, [url1, url2, disp])
return (
<mesh>
<planeBufferGeometry attach="geometry" args={[3.8, 3.8]} />
<shaderMaterial
attach="material"
args={[CrossFadeShader]}
uniforms-texture-value={texture1}
uniforms-texture2-value={texture2}
uniforms-disp-value={dispTexture}
uniforms-dispFactor-value={0.5}
/>
</mesh>
)
}
We support portals. You can use them to teleport a piece of the view into another container. Click here for a small demo.
import { createPortal } from 'react-three-fiber'
function Component() {
// "target" can be a three object, like a group, etc
return createPortal(<mesh />, target)
By default it renders like a game loop, which isn't that battery efficient. Switch on invalidateFrameloop
to activate loop invalidation, which is automatic most of the time.
<Canvas invalidateFrameloop ... />
Sometimes you must be able to kick off frames manually, for instance when you're dealing with async stuff or camera controls:
const { invalidate } = useThree()
const texture = useMemo(() => loader.load(url1, invalidate), [url1])
Supplying the vr
flag enables Three's VR mode and switches the render-loop to gl.setAnimationLoop as described in Three's docs.
import * as VR from '!exports-loader?WEBVR!three/examples/js/vr/WebVR'
import { Canvas } from 'react-three-fiber'
<Canvas vr onCreated={({ gl }) => document.body.appendChild(VR.createButton(gl))} />
If you want to exchange the default renderer you can. But, you will lose some of the functionality, like useRender, useThree, events, which is all covered in canvas. Here's a small example.
import { render, unmountComponentAtNode } from 'react-three-fiber'
const renderer = new THREE.SVGRenderer()
renderer.setSize(window.innerWidth, window.innerHeight)
document.body.appendChild(renderer.domElement)
const camera = new THREE.PerspectiveCamera(75, window.innerWidth / window.innerHeight, 0.1, 1000)
const scene = new THREE.Scene()
render((
<mesh>
<sphereGeometry name="geometry" args={[1, 16, 16]} />
<meshBasicMaterial name="material" />
</mesh>
), scene)
Three is a heavy-weight, and althought it is modular tree-shaking may not be sufficient. But you can always create your own exports file and alias the "three" name towards it. This way you can reduce it to 50-80kb or perhaps less.
Gist: https://gist.github.com/drcmda/974f84240a329fa8a9ce04bbdaffc04d