At Google I/O '24, we unveiled a vision of Android reimagined with AI at its core. Find the top 3 updates for building with AI on Android for Android developers in this video.
Choose the generative AI solution that's right for you
This document gives an overview of generative AI on Android, including available Gemini models and associated SDKs.
High performance on-device AI
On supported Android devices you can deliver rich generative AI experiences without needing a network connection or moving data off-device. On-device generative AI models such as Gemini Nano are great solutions for use-cases where low latency, low cost, and privacy safeguards are your primary concerns.
Use cases
- AI-enhanced content consumption: text summarization, document question answering, and entity extraction.
- AI-assisted content generation: proofreading, grammar correction, writing assistance, and contextual smart replies.
- Classifying text: sentiment or mood analysis
- Privacy: unlocks generative AI features while keeping data on-device
Solutions
Use the Google AI Edge SDK to leverage Gemini Nano inference on-device. Gemini Nano is now publicly available for experimental access.
Deliver custom models for on-device AI features more efficiently with Play for On-device AI. Google Play simplifies launching, targeting, versioning, downloading, and updating your on-device models, helping you to improve user experience while keeping your app's size optimized. Play for On-device AI is available at no extra cost. Complete the form if you are interested in Play for On-device AI early access.
Sign up for Play for On-device AI early access
Multimodal Cloud AI with Google's most capable models
You can create multimodal generative AI experiences in your apps by taking advantage of foundation models running inference in the cloud like the Gemini Pro models. These models are a great solution when you want to support the widest possible range of Android devices.
Use cases
- Image and video description and captioning: identifying objects and describing them in text
- Multimodal reasoning: processing text, image, and video content
- Text generation: summarize articles, answer questions about textual content, extract entities.
- Response formatting: format the model response to JSON or Markdown
Solutions
To bring your AI experiences into production, use Vertex AI in Firebase. The Firebase SDK provides access to Gemini models but also offers security and configuration options that are critical for production apps. In addition, Firebase includes production-level support and features across various mobile platforms.
Learn more about Vertex AI in Firebase
AI for Enterprises
Vertex AI is Google's fully-managed, unified AI development platform for AI. Enterprises can use Google's Vertex AI platform to deliver customized AI experiences to Android devices using backend integrations. Build, train, and deploy AI applications on Google's scalable, world-class infrastructure. It's a great solution for enterprise-scale AI, with access to over 130 models and tools including AI Studio, Agent Builder, and Gemini models.
Use cases
- Custom model training and delivery
- Image and video generation
- Virtual agents, customer support
- Speech to text, natural language processing
Solution
Use Google's Vertex AI platform to build custom AI applications and connect Android apps to the service layer.
Additional resources
Responsible Generative AI Toolkit
AI models should be aligned with safety policies, evaluated for fairness and accuracy, and designed to be transparent. The Responsible Generative AI Toolkit provides help and guidance for you to design, build, evaluate and deploy open AI models responsibly.
Gemini in Android Studio
Gemini in Android Studio is a coding companion for Android development. It's powered by artificial intelligence and can understand natural language. It helps you be more productive by answering your Android development queries. Gemini can help you find relevant resources, learn best practices, and save time.
Google APIs and SDKs for generative AI on Android
The following table is a brief overview of supported models for Vertex AI in Firebase and their latest stable model names. This table also lists preview and experimental models that are available for prototyping use cases. For additional details on each model's capabilities, including token and rate limits, see Gemini models.
| Model | Input | Output | Description |
|---|---|---|---|
| Gemini models with stable versions | |||
Gemini 2.0 Flashgemini-2.0-flash-001
|
text, code, PDFs, images, video, audio |
text, code, JSON (images & audio coming soon!) |
Provides next generation features and speed for a diverse variety of
tasks (multimodal generation coming soon!) |
Gemini 2.0 Flash‑Litegemini-2.0-flash-lite-001
|
text, code, PDFs, images, video, audio | text, code, JSON | Provides cost effective and low latency performance; supports high throughput |
Gemini 1.5 Progemini-1.5-pro-002
|
text, code, PDFs, images, video, audio | text, code, JSON | Supports complex reasoning tasks requiring more intelligence; 2M long context |
Gemini 1.5 Flashgemini-1.5-flash-002
|
text, code, PDFs, images, video, audio | text, code, JSON | Offers fast and versatile performance across a diverse variety of tasks |
| Gemini models with only preview and experimental versions (recommended for prototyping use cases only) | |||
Gemini 2.0 Progemini-2.0-pro-exp-02-05
|
text, code, PDFs, images, video, audio | text, code, JSON | Offers the strongest model quality, especially for code and world knowledge; 2M long context |
Gemini 2.0 Flash‑Thinking
gemini-2.0-flash-thinking-exp-01-21
|
text, code, PDFs, images | text, code, JSON | Offers stronger reasoning capabilities and includes the thinking process in responses |
| Imagen 3 models (when using with Vertex AI in Firebase) | |||
Imagen 3imagen-3.0-generate-002
|
text | images | Generates realistic, high-quality images from natural language text prompts |
Imagen 3 Fastimagen-3.0-fast-generate-001
|
text | images | Generates images for prototyping or low-latency use cases |