SceneView: An Open-Source 3D and AR SDK for Jetpack Compose Continuing Google Sceneform
Discover SceneView, an open-source 3D (Filament) and AR (ARCore) SDK built for Android Jetpack Compose and the web without heavy game engines. Explore glTF/USDZ
Adding interactive 3D rendering or Augmented Reality (AR) capabilities to mobile applications historically required embedding monolithic game engines like Unity or Unreal Engine, or wrestling directly with low-level graphics APIs. Following Google's discontinuation of Sceneform, the Android ecosystem lacked a maintained, lightweight graphics toolkit that seamlessly aligned with native UI architecture. SceneView addresses this gap as an open-source 3D and AR SDK built for modern Android Jetpack Compose, Compose Multiplatform, and the web, acting as the community-driven successor to Sceneform.

Image source: https://github.com/sceneview/sceneview
Distributed under the Apache License 2.0, SceneView relies on Google's high-performance Filament physically based rendering (PBR) engine and ARCore for spatial tracking. It enables engineering teams to embed interactive models and room-scale AR experiences directly into native mobile view hierarchies without bundling heavyweight external runtimes.
The Discontinuation of Google Sceneform and the Architecture of SceneView
When Google halted active development on Sceneform, development teams faced significant hurdles maintaining lightweight AR features, such as product previews and virtual furniture placement. SceneView fills this void by preserving Sceneform's familiar node-based scene graph structure while modernizing its architectural underpinnings for today's Android platform.
At its core, SceneView leverages Google Filament, a real-time mobile-first graphics engine widely used within Android itself. Filament delivers realistic material shading—including metallic properties, roughness, and image-based lighting reflections—while maintaining a modest memory and battery footprint. SceneView provides an ergonomic high-level wrapper around Filament, abstracting intricate graphics pipeline configurations into clean, declarative UI components.
Asset support is built on open standards. SceneView natively parses and renders glTF (GL Transmission Format) and GLB binary files, the industry standard for 3D on the web and mobile devices. It also supports USDZ (Universal Scene Description Zipped) models commonly found in Apple workflows. This dual support allows cross-functional teams to deploy shared 3D assets across platforms without recurring model conversion pipelines.
Declarative 3D and AR Workflows Built for Jetpack Compose
SceneView's primary strength lies in its seamless synchronization with Jetpack Compose's reactive state management paradigms. Traditional imperative rendering code requires developers to manually orchestrate lifecycle listeners, render loops, and view redraws. In SceneView, the 3D scene responds predictably to standard Kotlin state mutations.
To render a 3D scene, developers invoke the SceneView composable function, which automatically configures orbit camera gestures and balanced default lighting. Developers load assets asynchronously using the rememberModelInstance utility pointing to local bundle directories (models/helmet.glb). Once asset parsing completes, Compose triggers a recomposition that mounts the resulting ModelNode into the scene hierarchy. Because Filament graphics calls must execute strictly on the Android main thread, SceneView encapsulates these threading constraints internally, safeguarding against coroutine race conditions and memory leaks.
Augmented reality workflows follow the exact same declarative pattern through the arsceneview module. The ARSceneView composable pairs the camera stream with active ARCore tracking sessions. As camera frames detect horizontal or vertical surfaces, developers capture plane hits and store generated Anchor points inside Kotlin's mutableStateOf containers. The moment anchor state updates, Compose recomposes the tree, affixing the 3D model to physical space at a precise 1:1 real-world scale.
AI Assistant Readiness with MCP and llms.txt Integration
Reflecting emerging developer tooling workflows, SceneView introduces an explicit "AI-first" design philosophy. Large Language Models (LLMs) powering coding assistants like Claude Code and Cursor often hallucinate obsolete Sceneform methods or incorrect boilerplate when prompted for mobile 3D rendering code.
To eliminate this friction, the repository supplies structured reference guidelines directly within llms.txt and AGENTS.md. Furthermore, the project publishes sceneview-mcp, a dedicated Model Context Protocol server distributed via npm.
By registering this MCP server within their local agent environment, developers enable their AI assistants to inspect verified composable signatures, strict build constraints, and proper node hierarchy patterns in real time. The project team explicitly treats agent code generation failures as actionable defects: if an assistant fails to generate operational SceneView code on its first pass, the SDK maintains that the API design or accompanying documentation requires refinement.
Production Adoption, Toolchain Requirements, and Deprecation Boundaries
SceneView has advanced beyond an experimental utility into enterprise production validation. Boasting over 1,300 GitHub stars and recognized as a declared dependency across more than 500 public repositories, it powers features in major Google Play applications, including IKEA, Lenskart, Medal, and Safie. It is also integrated into Block's open-source Bitkey hardware wallet app and Niantic's Spatial SDK samples.
Engineering teams preparing to adopt the latest SceneView branch (4.69.x) must audit their build environments against its stringent toolchain requirements:
- Mandatory Modern Toolchain: Target application modules must configure
compileSdk = 37,minSdk = 24or higher, Android Gradle Plugin (AGP) 9.4.0 or above, Kotlin 2.3+, and a JVM compilation target of 21 across both Java and Kotlin. These strict constraints stem from SceneView shipping Java 21 bytecode alongside public inline functions. - Legacy Compatibility Baseline: Projects locked to legacy pipelines—such as AGP 8, compileSdk 36, or Kotlin 2.2—cannot resolve the latest release artifacts directly. These environments must pin their build configurations to
4.18.0, the final release compatible with earlier toolchains. - Platform Support Realignments: Active development is concentrated on native Android (Jetpack Compose), the web (
sceneview-webusing Filament.js), and Kotlin/Compose Multiplatform (sceneview-compose). Conversely, bridge packages for SwiftUI (SceneViewSwift), Flutter (flutter_sceneview), and React Native (@sceneview-sdk/react-native) have been formally deprecated as of version 4.68.0. While their source repositories remain accessible, they will no longer receive upstream updates.
For native Android teams looking to deliver interactive 3D visualization or practical AR placement without taking on the footprint of full-scale game engines, SceneView provides a modern, declarative graphics toolkit aligned with Android's current architecture.
Sources
- SceneView Official GitHub Repository: sceneview/sceneview
- Stellargent Lab Post on X (@stellargent_lab): 2026-11-15 SceneView Open-Source SDK Introduction