Custom Models

If AutoML or the base APIs in ML Kit don’t cover your use cases, you can bring your own existing TensorFlow Lite models. Just upload your model to ML Kit and we’ll take care of hosting and serving it to your app.


Smaller app install size

Dynamically download ML models only when needed, or bundle with the app to be available at install time.

Regular model updates

Iterate, update, and serve models to your users without having to resubmit a new version of your app.

A/B testing

Easily experiment with multiple models to see which performs best using Firebase A/B testing.

Dynamic model selection

Simultaneously and dynamically choose which model to serve to a specific user segment with Firebase RemoteConfig.

AutoML Vision Edge

If you need a more specialized version of the image labeling API, covering a narrower domain of concepts in more detail (for example, species of flowers or types of food), AutoML Vision Edge can be used to easily train a custom on-device model with your own images.

TensorFlow Lite

Bundling a model directly with TensorFlow Lite is also an easy alternative in cases where model serving, experimentation, and updates aren't essential to your app.