Google MediaPipe is a cross-platform, on-device machine learning framework enabling highly optimized real-time vision, text, audio, and generative AI inference.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install mediapipe
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install mediapipe using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Google MediaPipe?
Google MediaPipe is an open-source framework designed for building state-of-the-art, on-device machine learning pipelines. Developers can deploy highly efficient solutions across Android, iOS, web, desktop, and IoT devices. The framework operates on two distinct levels: high-level MediaPipe Solutions, which offer pre-built task APIs with optimized models, and low-level MediaPipe Framework, a graph-based pipeline builder designed for custom C++ workloads.
Utilizing MediaPipe via Openclaw Skills empowers developers to incorporate low-latency tracking, vision processing, and voice controls into their automated agents without resorting to high-latency external cloud APIs. From running facial landmark detection to executing local quantized large language models, MediaPipe is engineered to maximize performance directly on edge hardware.
Google MediaPipe Use Cases
- Creating low-latency computer vision solutions including real-time face tracking, body pose estimation, and gesture controls.
- Performing offline-first text analysis, sentiment classification, language identification, and semantic document comparison.
- Implementing real-time on-device audio classification from live microphonic inputs or packaged audio clips.
- Deploying private, hardware-accelerated on-device generative AI models and text chat capabilities.
- Fine-tuning pre-trained image classifiers or object detectors using transfer learning on small custom datasets.
How Google MediaPipe Works
- Retrieve or customize a pre-trained task bundle (.task or .tflite format) from the official model library.
- Initialize your target task configuration, specifying either IMAGE, VIDEO, or LIVE_STREAM execution modes.
- Configure the hardware acceleration delegate to utilize CPU, GPU, or mobile NPU processing dynamically.
- Pack raw input frames or buffers into typed data packets recognized by the underlying graph structure.
- Run model inference synchronously or register an asynchronous event callback handler for live camera streams.
- Retrieve formatted landmark coordinates, bounding boxes, masks, or classifications for downstream app logic.
Google MediaPipe Setup
Python Integration
Install the main library package:
pip install mediapipe
If you want to fine-tune existing models using custom training data:
pip install mediapipe-model-maker
Web Deployment
To use MediaPipe in web applications, install the task-specific packages:
npm install @mediapipe/tasks-vision
Android Build Configuration
Add the vision task library to your build.gradle file:
implementation 'com.google.mediapipe:tasks-vision:0.10.35'
iOS Dependency Management
Include the pod file dependency in your local configuration:
pod 'MediaPipeTasksVision'
Google MediaPipe Data Schema & Taxonomy
MediaPipe uses standardized model bundles and structured pipeline schemas to represent input resources and processing graphs.
Vision Task Schema Mapping
| Vision Task | Model Asset File | Primary Output Struct | Configurable Options |
|---|---|---|---|
| FaceLandmarker | face_landmarker.task |
478 3D landmarks, blendshape scores, transform matrices | num_faces, output_face_blendshapes |
| HandLandmarker | hand_landmarker.task |
21 hand landmarks, handedness classification, world coordinates | num_hands, min_tracking_confidence |
| PoseLandmarker | pose_landmarker_lite.task |
33 pose landmarks, 3D world landmarks, segmentation mask | num_poses, output_segmentations |
| HolisticLandmarker | holistic_landmarker.task |
Combined landmark lists (face, hands, pose) simultaneously | min_face_detection_confidence |
| GestureRecognizer | gesture_recognizer.task |
Landmarker output with predefined gestural categories | min_hand_detection_confidence |
Low-Level Pipeline Streams
- Packets: Strongly typed wrapper components carrying underlying hardware-mapped frames, float arrays, or landmarks.
- Streams: Structured flow channels defining inputs, outputs, and logical connections within the pipeline graph.
- Graphs (.pbtxt): Plain text protobuffer definitions configuration specifying calculators, stream topologies, and options.
Google MediaPipe Advanced Features
- Local LLM Inference: Executes lightweight generative AI models on-device using quantized weights through web browsers or local mobile NPUs.
- Real-Time Live Streaming: Features low-latency, callback-driven pipelines for processing live video and audio inputs asynchronously.
- Custom Calculator Nodes: Allows C++ developers to construct custom data processing calculators to build custom pipeline architectures.
- Hardware-Aware Delegation: Easily toggles processing paths between local CPU, GPU, or specialized neural execution engines.
- Transfer Learning API: Seamlessly integrates with MediaPipe Model Maker to customize models using few-shot training examples.
SKILL.md
Loading
Related Openclaw Skills
Image Watermark
An automated utility for processing and adding custom text watermarks to single or multiple images simultaneously.
Claude Code Agent SDK
A developer-centric SDK to build, configure, and deploy production-grade AI agents with Claude Code as a library in Python and TypeScript.
Daily News Auto-Collector & Report Generator
An automated news aggregation tool that leverages AI-driven web search and 24-hour filters to generate noise-free daily briefings.
Daily News Automated Collector & Reporter v2.0
An AI-powered automated daily news collection and synthesis agent skill utilizing strict 24-hour freshness web search filtering.
bee-languages-get
A high-performance retrieval tool for obtaining a list of active site languages from the Bee Website Builder API to power dynamic localization workflows.
Facebook Page Ronnie
Automate your Facebook Page publishing and audience engagement via robust Graph API connections and intelligent browser fallback mechanisms.
Featured*
