Get and build the SDK

Use a prebuilt SDK when its platform, backend and API version fit your application. Build from source when you need newer interfaces, a different toolchain or custom backend options. This section covers the native libraries first, then packaging them for an application or Python environment.

Prebuilt SDKs

The download list below reflects GitHub Releases on September 29, 2026. The latest published native SDK is v1.2.3. Links point to that release so the selected package does not change underneath an integration.

The CUDA/Ubuntu label above is the release asset name. Check the linked library dependencies on the deployment machine. The release has no separate HarmonyOS, Windows or CoreML archive; HarmonyOS and the Apple build pages cover their source build routes. There is no dedicated Windows build guide in this section.

Choose the library for the application process, including its architecture and C runtime. A 64-bit device can still run a 32-bit application. Keep each archive’s headers and libraries together, and check the platform guide before linking static frameworks or GPU/NPU libraries.

Apple SDK packaging

The 1.2.4 source adds Objective-C and Swift APIs, iOS simulator slices and paired XCFrameworks. The v1.2.3 downloads above use the older package layout; follow Develop source setup to build the SDK for the Apple examples in this documentation.

InspireFace.framework contains the C and Objective-C interfaces. InspireFaceSwift.framework adds Swift configuration types and scoped buffer helpers. Keep both frameworks from the same build. The package also includes merged platform frameworks, native SDK directories and metadata for checking architectures and deployment targets.

Use the iOS build guide for device and simulator packages, and the macOS build guide for Intel, Apple Silicon and universal packages. iOS frameworks are static; macOS frameworks are dynamic. This distinction controls Xcode's embedding settings. The new iOS framework already incorporates its inference dependency, so it does not need an additional MNN.framework in the application target.

The Apple release workflow prepares a CPU archive named inspireface-apple-<version>.zip; no such archive is present in the published release checked above. CoreML packages can be built locally with the same builder and require a compatible CoreML resource pack.

Python and Android packages

python -m pip install inspireface opencv-python

For a custom native library or wheel, see Python packaging and library replacement.

The Android example uses the JitPack dependency com.github.HyperInspire:inspireface-android-sdk:1.2.0. Its integration page includes the repository and Gradle configuration. Building a native Android SDK produces JNI libraries; assembling Java classes and libraries into an AAR is a separate packaging step explained in Android builds.

Download the model separately

SDK libraries and Python wheels need a resource pack at runtime. Download packs from the model release: Pikachu and Megatron for CPU, Megatron_TRT for TensorRT, or the Gundam file for the target Rockchip SoC. See model selection and loading for the full mapping.

Choose a build guide