MediaPipe Tasks provides cross-platform APIs and prebuilt pipelines for running vision, text, audio, retrieval, and decision models directly on-device. Every task pairs an end-to-end runtime pipeline with hardware-accelerated execution across Web, Android, iOS, and Python. To test tasks interactively before writing code, launch [MediaPipe Web Demos](https://google-ai-edge.github.io/mediapipe-samples-web/), browse [Samples \& Demos](https://developers.google.com/edge/mediapipe/demos), or jump directly to any task guide for a live Web SDK playground and platform **Quickstart** tabs.  Consistent API Three steps across platforms Install the domain package, initialize the task with a `.task`, `.tflite`, or `.litertlm` model bundle, and run inference on images, video frames, text, or audio streams. Hardware Acceleration End-to-end CPU and GPU pipelines MediaPipe Tasks handles input preprocessing, tokenization, model execution, and structured post-processing on-device for low-latency real-time applications. ## Model asset placement \& threading Every MediaPipe task loads a `.task` bundle, `.tflite` model, or `.litertlm` checkpoint via `BaseOptions`. Place your model file in the standard asset location for your target platform: - **Android** : Place `.task`, `.tflite`, or `.litertlm` files in your module's `src/main/assets/` directory (for example, `app/src/main/assets/hand_landmarker.task`) and pass the filename to `BaseOptions.builder().setModelAssetPath("hand_landmarker.task")`. - **iOS** : Add the model file to your Xcode application target bundle (verified under **Build Phases \> Copy Bundle Resources** ) and resolve its runtime path with `Bundle.main.path(forResource:ofType:)`. - **Web** : Serve the model from your application's static public directory or a CDN URL and pass the URL to `baseOptions.modelAssetPath` (or pass a `Uint8Array` / `ReadableStreamDefaultReader` to `baseOptions.modelAssetBuffer`). - **Python** : Pass the local filesystem path to `BaseOptions(model_asset_path="...")`. > [!NOTE] > **Note:** **Threading and UI responsiveness:** Synchronous inference calls in `IMAGE`, `VIDEO`, and `AUDIO_CLIPS` running modes block the calling thread until execution completes. To avoid blocking the UI thread, call synchronous task methods from a background executor (`ExecutorService` or Kotlin coroutine on Android, a background `DispatchQueue` on iOS, or a Web Worker on Web). On iOS, `LIVE_STREAM` and `AUDIO_STREAM` delegate callbacks are invoked on a dedicated serial background queue---dispatch any UI updates back to `DispatchQueue.main.async`. ## Supported platforms Choose a platform to add the MediaPipe Tasks SDK to your project across all five task domains (**Vision** , **Text** , **Audio** , **Retrieval** , and **Decision** ), read the complete setup guide for each environment, and see [Prepare input data](https://developers.google.com/edge/mediapipe/solutions/prepare_data) to format vision and audio media for inference. ### Web and JavaScript Install the npm package for your domain (`@mediapipe/tasks-vision`, `@mediapipe/tasks-text`, `@mediapipe/tasks-audio`, `@mediapipe/tasks-retrieval`, or `@mediapipe/tasks-decision`) or import its ESM bundle from a CDN. See the [Web setup guide](https://developers.google.com/edge/mediapipe/solutions/setup_web) and [JavaScript API reference](https://developers.google.com/edge/api/mediapipe/js/tasks-vision). npm install @mediapipe/tasks-vision @mediapipe/tasks-text @mediapipe/tasks-audio @mediapipe/tasks-retrieval @mediapipe/tasks-decision ### Android Add the Maven dependency for your domain in Gradle. See the [Android setup guide](https://developers.google.com/edge/mediapipe/solutions/setup_android) and [Java API reference](https://developers.google.com/edge/api/mediapipe/java/com/google/mediapipe/tasks/core/package-summary). ### Android (Kotlin) dependencies { implementation("com.google.mediapipe:tasks-vision:latest.release") implementation("com.google.mediapipe:tasks-text:latest.release") implementation("com.google.mediapipe:tasks-audio:latest.release") implementation("com.google.mediapipe:tasks-retrieval:latest.release") implementation("com.google.mediapipe:tasks-decision:latest.release") } ### Android (Java) dependencies { implementation 'com.google.mediapipe:tasks-vision:latest.release' implementation 'com.google.mediapipe:tasks-text:latest.release' implementation 'com.google.mediapipe:tasks-audio:latest.release' implementation 'com.google.mediapipe:tasks-retrieval:latest.release' implementation 'com.google.mediapipe:tasks-decision:latest.release' } ### iOS Add MediaPipe Tasks using Swift Package Manager in Xcode or `Package.swift`. See the [iOS setup guide](https://developers.google.com/edge/mediapipe/solutions/setup_ios) and [Swift API reference](https://developers.google.com/edge/api/mediapipe/swift/vision/Classes). // Package.swift (Swift Package Manager) dependencies: [ .package(url: "https://github.com/google-ai-edge/mediapipe", branch: "master"), ], targets: [ .target( name: "MyAppTarget", dependencies: [ .product(name: "MediaPipeTasksVision", package: "mediapipe"), .product(name: "MediaPipeTasksText", package: "mediapipe"), .product(name: "MediaPipeTasksAudio", package: "mediapipe"), .product(name: "MediaPipeTasksRetrieval", package: "mediapipe"), .product(name: "MediaPipeTasksDecision", package: "mediapipe"), ] ), ] ### Python Install the `mediapipe` package with `pip` and import the task modules for desktop, server, or notebook workflows. See the [Python setup guide](https://developers.google.com/edge/mediapipe/solutions/setup_python) and [Python API reference](https://developers.google.com/edge/api/mediapipe/python/mp). # python -m pip install mediapipe import mediapipe as mp from mediapipe.tasks import python from mediapipe.tasks.python import audio, decision, retrieval, text, vision