ML Kit’s Prompt API: Unlock Custom On-Device Gemini Nano Experiences

Today marks a major milestone for Android's on-device generative AI. We're announcing the Alpha release of the ML Kit GenAI Prompt API. This API allows you to send natural language and multimodal requests to Gemini Nano, addressing the demand for more control and flexibility when building with generative models.

Partners like Kakao are already building with Prompt API, creating unique experiences with real-world impact. You can experiment with Prompt API's powerful features today with minimal code.

Move beyond pre-built to custom on-device GenAI Prompt API moves beyond pre-built functionality to support custom, app-specific GenAI use cases, allowing you to create unique features with complex data transformation. Prompt API uses Gemini Nano on-device to process data locally, enabling offline capability and improved user privacy.

Key use cases for Prompt API:

Prompt API allows for highly customized GenAI use cases. Here are some recommended examples: 

• Image understanding: Analyzing photos for classification (e.g., creating a draft social media post or identifying tags such as "pets," "food," or "travel").

• Intelligent document scanning: Using a traditional ML model to extract text from a receipt, and then categorizing each item with Prompt API.

• Transforming data for the UI: Analyzing long-form content to create a short, engaging notification title.

• Content prompting: Suggesting topics for new journal entries based on a user’s preference for themes.

• Content analysis: Classifying customer reviews into a positive, neutral, or negative category.

• Information extraction: Extracting important details about an upcoming event from an email thread.

Prompt API lets you create custom prompts and set optional generation parameters with just a few lines of code:

Generation.getClient().generateContent(
   generateContentRequest(
       ImagePart(bitmapImage),
       TextPart("Categorize this image as one of the following: car, motorcycle, bike, scooter, other. Return only the category as the response."),
   ) {
       // Optional parameters
       temperature = 0.2f
       topK = 10
       candidateCount = 1
       maxOutputTokens = 10
   },
)

For more detailed examples of implementing Prompt API, check out the official documentation and sample on Github.

Gemini Nano, performance, and prototyping

Prompt API currently performs best on the Pixel 10 device series, which runs the latest version of Gemini Nano (nano-v3). This version of Gemini Nano is built on the same architecture as Gemma 3n, the model we first shared with the open model community at I/O.

The shared foundation between Gemma 3n and nano-v3 enables developers to more easily prototype features. For those without a Pixel 10 device, you can start experimenting with prompts today by prototyping with Gemma 3n locally or accessing it online through Google AI Studio.

For the full list of devices that support GenAI APIs, refer to our device support documentation.

Learn more