Gemini Nano on device
summary
Gemini Nano weighs 3–4 GB so it ships once in the system and every app shares it via AI Core; foreground apps get top priority, background jobs queue and run overnight on charge. GenAI MLKit APIs require flagship devices from the last couple of years while classic MLKit runs on a billion+ devices. Hybrid inference falls back to Gemini Flash in the cloud; an embedding API is coming soon and LiteRT is the other path.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:30On-device AI options
- 3:18Gemini Nano and AI Core
- 5:01Gemini Nano via ML Kit GenAI
- 6:07Hybrid inference with Firebase AI
- 7:29Android AI offerings overview
- 8:11Q&A on RAM and battery
- 10:17Model sharing and optimization
- 12:49Local vs remote inference
- 15:18Building RAG with Prompt API
- 16:51System prompts and file access
- 17:56Device and model compatibility
related talks
41mBuild intelligent Android apps with Google's AI
10mAI in your pocket: Building intelligent Android apps
20mWhat's new in Machine Learning Keynote
23m2025: On‑Device AI Goes Mainstream on Android
46mSmall AI is the next Big Thing
35mOn‑Device LLMs on Android Devices 2025
18mLive demo: Gemma 3 on Jetson Orin Nano
11mapproach to generative AI