Generative AI advancements are setting the stage for a major change in mobile technology. Google aims to support developers by providing tools that utilize on-device AI capabilities, despite most current AI features running on powerful remote servers. Google’s forthcoming I/O event has generated heightened expectations as industry observers believe the company will reveal new APIs to integrate the Gemini Nano model’s capabilities directly onto Android phones. The initiative enables advanced AI features to reach users directly while boosting privacy and potentially speeding up operations through reduced cloud processing dependency.

New information found in Google’s developer documentation provides insights into upcoming AI features. Android Authority has announced that the upcoming ML Kit SDK update will bring API support for Gemini Nano-powered on-device generative AI capabilities. The framework combines AI Core elements from the Edge AI SDK but simplifies development with existing model integration and clearly defined features for developers. The update implies that developers of various skill levels will find practical tools to easily incorporate AI functionality into their mobile apps.

Google’s documentation specifies that the new ML Kit GenAI APIs permit applications to execute multiple essential tasks on-device without sending sensitive user data to the cloud. The API supports multiple functions such as text summarization alongside proofreading, rewriting capabilities, and image description features. The processing power limitations inherent to mobile devices create specific operational limits for the on-device version of Gemini Nano. Summaries will not exceed three bullet points, and initial image descriptions will only be provided in English. The quality of AI outputs differs from phone to phone based on the Gemini Nano version installed. While the standard Gemini Nano XS occupies about 100MB of space, the Gemini Nano XXS version, which powers devices such as the Pixel 9a, occupies only 25MB and functions with text-only output and a reduced context window.

Google’s strategic decision will benefit the entire Android community because the ML Kit SDK works with devices that aren’t limited to the Pixel series. Pixel phones make extensive use of Gemini Nano while manufacturers such as OnePlus with their 13 model, Samsung with the Galaxy S25, and Xiaomi with their 15 device are also building their phones to support this AI model. By integrating Google’s on-device AI model into more Android phones, developers can reach broader audiences with generative AI-powered features, which will promote innovation and enhance user-friendly mobile experiences across different brands.

Android app developers aiming to implement on-device generative AI features face limited available options. The Neural Processing Unit (NPU) access within Google’s experimental AI Edge SDK runs AI models but remains limited to the Pixel 9 series and text processing only. Using Qualcomm and MediaTek APIs for AI workloads presents a risk to long-term projects because their features and capabilities differ greatly across devices. Implementing custom AI models demands a deep understanding of generative AI systems. The new APIs will streamline local AI implementation so developers can work faster and more easily across a broader spectrum.

On-device AI models have their limitations when compared to cloud-based systems, but this advancement represents an important step forward in making artificial intelligence more seamlessly integrated into everyday life. The majority of users will likely choose to process their data locally to gain enhanced privacy and security instead of transmitting it to distant servers. Google’s Pixel Screenshots processes images directly on the device, while Motorola’s high-end Razr Ultra implements local notification summarization, unlike the base Razr model, which uses cloud processing to demonstrate the advantages of on-device processing. Standardized APIs centered around Gemini Nano can deliver much-needed uniformity to mobile AI development. The effectiveness of Gemini Nano depends on how Google and other Original Equipment Manufacturers (OEMs) work together to support its implementation on various Android devices, because certain companies might adopt different methods, while older or lower-spec phones may not support local AI processing.