MediaPipe

An open-source on-device AI toolkit for vision, text and audio across mobile, web and desktop, with pretrained models and customization tools.

MediaPipe is a toolkit for developers adding on-device AI to applications. It provides pretrained models and task APIs for applications that need to process images, video, text or audio on the user's device.

Vision tasks: Detect objects, track face, hand and pose landmarks, recognize gestures, segment images, or classify and compare images through embeddings.
Text and audio: Add audio classification, language detection, text classification and embeddings, proofreading or summarization; an LLM Inference API is also available.
Model customization: Use Model Maker to adapt supported models with your own data, and MediaPipe Studio to inspect results, evaluate solutions and benchmark them in a browser.
Custom pipelines: MediaPipe Framework provides a lower-level foundation for developers who need processing pipelines beyond the supplied tasks.

MediaPipe is open source under Apache 2.0. It supports Android, iOS, web, desktop and edge applications, with Python APIs also available. Task availability and model customization support vary by platform and solution.

MediaPipe Tasks processes input on-device and doesn't send that input to Google servers. Its APIs do send performance and usage metrics to Google. Legacy solutions whose support has ended remain available as-is.