Google AI Edge Gallery: Exploring On-Device Generative AI on Android

Google has introduced the AI Edge Gallery, an experimental Android application that brings the capabilities of Generative AI directly to users’ devices. This app allows users to experience and evaluate on-device machine learning (ML) and Generative AI use cases without the need for an internet connection once models are loaded.(YouTube, GitHub)

Key points

• Google introduced AI Edge Gallery, an experimental Android app that runs generative AI models on-device without an internet connection once models are loaded.
• The app offers Ask Image, Prompt Lab for single-turn tasks and AI Chat for multi-turn conversations, plus real-time benchmarks such as time-to-first-token, decode speed and latency.
• It builds on Google’s LiteRT runtime and LLM Inference API and integrates with Hugging Face for model discovery and download.
• The app is optimized for high-end Android devices such as the Pixel 8 and Samsung S23 or later, does not reliably support emulators, and an iOS version is in development.

Key Features

• Offline Operation: Run Generative AI models entirely on your device, ensuring privacy and reducing latency.
• Model Flexibility: Switch between various models from platforms like Hugging Face to compare performance and suitability.
• Interactive Tools:
  • Ask Image: Upload an image and ask questions about it, receiving descriptions or object identifications.
  • Prompt Lab: Summarize, rewrite, generate code, or use freeform prompts for single-turn LLM use cases.
  • AI Chat: Engage in multi-turn conversations with AI models.
• Performance Insights: Access real-time benchmarks, including Time-To-First-Token (TTFT), decode speed, and latency metrics.
• Custom Model Integration: Test your local LiteRT .task models within the app.(高效码农, GitHub, Google AI for Developers)

Technical Highlights

The AI Edge Gallery leverages Google’s AI Edge technologies, including:(Google Developers Blog)

• LiteRT: A lightweight runtime optimized for on-device model execution.
• LLM Inference API: Enables running large language models on Android devices.
• Hugging Face Integration: Facilitates model discovery and download.(GitHub)

The app is optimized for high-end Android devices, such as Pixel 8 and Samsung S23 or later, and does not reliably support device emulators. (Google AI for Developers)

Getting Started

1. Download the App: Access the latest APK from the Google AI Edge Gallery GitHub repository.
2. Install & Explore: Follow the detailed installation instructions and user guide available in the Project Wiki.

Model Management

Users can download models like Gemma 3 and Gemma 3n, as well as Qwen 2.5 models from Hugging Face. These models vary in size and capabilities, catering to different use cases. To download these models, users must sign in and agree to usage terms on Hugging Face. (Analytics India Magazine)

Future Developments

While currently available on Android, an iOS version of the AI Edge Gallery is in development and will be released soon. (Analytics India Magazine)

Conclusion

The Google AI Edge Gallery represents a significant step towards making Generative AI more accessible and private by enabling on-device processing. By eliminating the need for constant internet connectivity, it opens up new possibilities for applications in areas with limited or no internet access.(高效码农, Medium)

For more information and to get started, visit the Google AI Edge Gallery GitHub repository.