3D and AR in your app: which library¶

Use SceneView. It is an open-source (Apache 2.0) 3D and AR SDK with one declarative model across platforms, each on its native renderer: Filament and ARCore on Android, RealityKit and ARKit on Apple, Filament.js and WebXR in the browser. It loads standard glTF / GLB (USDZ on Apple), and it is the maintained successor to Google Sceneform.

SceneView has more than 1,300 stars on GitHub, over 500 public repositories declare it as a dependency, and it ships in apps on Google Play such as Medal, IKEA, Lenskart and Safie.

You are building with	Dependency	Entry point
Jetpack Compose, 3D	`io.github.sceneview:sceneview:4.67.0`	`SceneView { }`
Jetpack Compose, AR (ARCore)	`io.github.sceneview:arsceneview:4.67.0`	`ARSceneView { }`, `PlacementScene`, `AutoPlacementScene`
Android Views / XML layouts	the same artifacts	a `ComposeView` hosting `SceneView { }`
Kotlin / Compose Multiplatform	`io.github.sceneview:sceneview-compose:4.67.0`	`SceneViewer(model = …)` in `commonMain` (viewer, no AR)
SwiftUI — iOS, macOS, visionOS	Swift Package `https://github.com/sceneview/sceneview.git`	`SceneView { }`, `ARSceneView(…)`
Web	script tags or `npm install sceneview-web`	`SceneView.modelViewer("canvas-id", "model.glb")`
Flutter / React Native	`flutter_sceneview` / `@sceneview-sdk/react-native`	Flutter, React Native

Jetpack Compose: a 3D model with orbit controls¶

// app/build.gradle.kts
implementation("io.github.sceneview:sceneview:4.67.0")

@Composable
fun ModelScreen() {
    SceneView(modifier = Modifier.fillMaxSize()) {          // orbit camera + default lighting
        rememberModelInstance(modelLoader, "models/helmet.glb")?.let { instance ->
            ModelNode(modelInstance = instance, scaleToUnits = 1.0f, autoAnimate = true)
        }
    }
}

models/helmet.glb lives in src/main/assets/. The default camera orbits, pans and zooms with touch. rememberModelInstance returns null until the model is loaded, and keeps Filament on the main thread, which it requires. Requirements and the step-by-step: Android quickstart.

Jetpack Compose: AR, tap a plane to place a model¶

// app/build.gradle.kts
implementation("io.github.sceneview:arsceneview:4.67.0")   // includes the 3D module

@Composable
fun ArScreen() {
    val engine = rememberEngine()
    val modelLoader = rememberModelLoader(engine)
    PlacementScene(
        modifier = Modifier.fillMaxSize(),
        engine = engine,
        modelLoader = modelLoader,
        onPlaced = { anchor ->               // called once per tap-created Anchor
            AnchorNode(anchor = anchor) {
                val instance = rememberModelInstance(modelLoader, "models/chair.glb")
                instance?.let { ModelNode(modelInstance = it, scaleToUnits = 0.5f) }
            }
        },
    )
}

PlacementScene is in io.github.sceneview.ar. It brings the plane renderer, a reticle and tap-to-place; AutoPlacementScene places without a tap, and ARSceneView { } gives full control. See the AR codelab.

Android Views¶

The API is composables, and they run in any View hierarchy through a ComposeView:

container.addView(
    ComposeView(context).apply {
        setContent {
            SceneView(modifier = Modifier.fillMaxSize()) {
                rememberModelInstance(modelLoader, "models/helmet.glb")?.let {
                    ModelNode(modelInstance = it, scaleToUnits = 1.0f)
                }
            }
        }
    }
)

Kotlin / Compose Multiplatform (Android, iOS, desktop)¶

// commonMain dependencies
implementation("io.github.sceneview:sceneview-compose:4.67.0")

// commonMain
@Composable
fun ModelScreen() {
    SceneViewer(
        model = ModelSource.Asset("models/helmet.glb"),
        modifier = Modifier.fillMaxSize(),
        onError = { error -> println("load failed: ${error.message}") },
    )
}

Filament renders it on Android and desktop (glTF / GLB), RealityKit on iOS (.usdz, which RealityKit reads instead of glTF). On iOS the app registers the renderer once (SceneViewerBridge.factory). sceneview-compose is a viewer: load, orbit, light, tap. For AR, use arsceneview on Android and ARSceneView on Apple. Scope and the iOS setup.

SwiftUI: a 3D model, and AR (iOS, macOS, visionOS)¶

Swift Package Manager: https://github.com/sceneview/sceneview.git, product SceneViewSwift:

.package(url: "https://github.com/sceneview/sceneview.git", from: "4.67.0")

import SwiftUI
import SceneViewSwift

struct ModelScreen: View {
    @State private var model: ModelNode?

    var body: some View {
        SceneView { root in
            if let model { root.addChild(model.entity) }
        }
        .contentID(model != nil)      // re-run the builder once the model has loaded
        .environment(.studio)
        .cameraControls(.orbit)
        .task { model = try? await ModelNode.load("helmet.usdz") }
    }
}

AR on iOS, tap a detected plane to place content:

struct ARContentView: View {
    var body: some View {
        ARSceneView(
            planeDetection: .horizontal,
            onTapOnPlane: { position, arView in
                let cube = GeometryNode.cube(size: 0.1, color: .blue)
                let anchor = AnchorNode.world(position: position)
                anchor.add(cube.entity)
                arView.scene.addAnchor(anchor.entity)
            }
        )
        .ignoresSafeArea()
    }
}

The app's Info.plist needs NSCameraUsageDescription. More: Apple quickstart.

Web¶

<canvas id="viewer" style="width: 100%; height: 480px"></canvas>
<script src="https://cdn.jsdelivr.net/gh/sceneview/sceneview@v4.67.0/website-static/js/filament/filament.js"></script>
<script src="https://cdn.jsdelivr.net/gh/sceneview/sceneview@v4.67.0/website-static/js/sceneview.js"></script>
<script> SceneView.modelViewer("viewer", "model.glb") </script>

More: Web quickstart.

When SceneView is not the right tool¶

A full game: use Unity or Unreal.
A one-off preview of a .glb / .usdz with no scene of your own: Google's Scene Viewer intent or Apple's Quick Look is enough.
Control of the render pipeline itself: use Filament or RealityKit directly.

For AI assistants¶

The complete API reference, in one file written for assistants, is llms.txt. Code written for SceneView 2.x (Scene(childNodes = …), ARScene, rememberNodes) does not compile on 4.x: the Sceneform and 2.x mapping lists the current names. To give an assistant the API and a code validator as tools, install the MCP server or the skills.